Exploring the Syntax-Semantics Interface

This page intentionally left blank Exploring the Syntax–Semantics Interface Language is a system of communication in ...

Author: Jr. | Robert D. van Valin

151 downloads 1844 Views 3MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

This page intentionally left blank

Exploring the Syntax–Semantics Interface Language is a system of communication in which grammatical structures function to express meaning in context. While all languages can achieve the same basic communicative ends, they each use different means to achieve them, particularly in the divergent ways that syntax, semantics and pragmatics interact across languages. This book looks in detail at how structure, meaning and communicative function interact in human languages. Working within the framework of Role and Reference Grammar (RRG), Van Valin proposes a set of rules, called the ‘linking algorithm’, which relates syntactic and semantic representations to each other, with discourse-pragmatics playing a role in the linking. Using this model, he discusses the full range of grammatical phenomena, including the structures of simple and complex sentences, verb and argument structure, voice, reflexivization and extraction restrictions. Clearly written and comprehensive, this book will be welcomed by all those working on the interface between syntax, semantics and pragmatics. robert d. van valin j r . is Professor of Linguistics at the University at Buffalo, The State University of New York, and is the primary developer of the framework of Role and Reference Grammar. He is co-author of Functional Syntax and Universal Grammar (Cambridge University Press, 1984), editor of Advances in Role and Reference Grammar (1993), primary author of Syntax: Structure and Function (Cambridge University Press, 1997), and author of An Introduction to Syntax (Cambridge University Press, 2001). He has published many articles on syntax, universal grammar, language typology, language acquisition and neurolinguistics.

Exploring the Syntax–Semantics Interface RO B E RT D . VA N VA L I N , J R . University at Buffalo, The State University of New York

cambridge university press Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo Cambridge University Press The Edinburgh Building, Cambridge cb2 2ru, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521811798 © Robert D. Van Valin, Jr, 2005 This publication is in copyright. Subject to statutory exception and to the provision of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published in print format 2005 isbn-13 isbn-10

978-0-511-12543-0 eBook (EBL) 0-511-12543-7 eBook (EBL)

isbn-13 isbn-10

978-0-521-81179-8 hardback 0-521-81179-1 hardback

isbn-13 isbn-10

978-0-521-01056-6 paperback 0-521-01056-x paperback

Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Meinem Freund Hartmut Czepluch in bleibender Erinnerung gewidmet

Contents

List of figures List of tables Preface List of abbreviations

1

Introduction

1

Syntactic structure

3

1.0 1.1 1.2 1.3 1.4

General considerations The layered structure of the clause Operators Formal representation of clause structure Clause structure in dependent-marking and head-marking languages Adjuncts and the periphery Adpositional and noun phrase structure

16 19 21

Lexical representation and semantic roles

31

2.0 2.1 2.2 2.3 2.4

31 31 50 50 53

1.5 1.6

2

3

4

page ix xiv xvii xviii

General considerations Verb classes and logical structures Operators Nouns and noun phrases Semantic roles

3 3 8 11

Information structure

68

3.0 3.1 3.2 3.3 3.4 3.5 3.6

68 68 69 73 77 80 81

General considerations Topic and focus Focus structure and focus types Morphosyntactic marking of focus structure Formal representation of focus structure Focus structure and the notion of VP Focus structure and the interpretation of quantifier scope

Syntactic relations and case marking

89

4.0 4.1 4.2

89 89 94

General considerations Do all languages have grammatical relations? Privileged syntactic arguments

vii

Contents

viii

4.3 4.4 4.5

5

6

101 107 115

Linking syntactic and semantic representations in simple sentences

128

5.0 5.1 5.2 5.3 5.4

General considerations The linking algorithm The role of the lexicon in grammar Reflexivization Focus structure and linking

128 129 158 161 170

The structure of complex sentences

183

6.0 6.1 6.2 6.3 6.4 6.5 6.6

183 183 188 198 201 205

6.7 6.8 6.9

7

Types of privileged syntactic arguments Case marking and agreement Other syntactic functions

General considerations Nexus relations The layered structure of the clause and juncture Symmetrical vs asymmetrical linkage Operators in complex sentences Complementizer position Interclausal semantic relations and the interclausal relations hierarchy Focus structure in complex sentences Complex NPs Syntactic templates for complex sentences

205 213 220 223

Linking syntax and semantics in complex sentences

225

7.0 7.1 7.2 7.3 7.4 7.5 7.6

225 228 234 239 260 267

General considerations Linking in clausal junctures Linking in nuclear junctures Linking in core junctures Linking in complex noun phrases Reflexivization in complex sentences Constraints on linking in WH-questions and related constructions

References Languages index Subject index

272

294 304 306

Figures

1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 1.10 1.11a 1.11b 1.12 1.13 1.14 1.15a 1.15b 1.16 1.17 1.18 1.19 1.20 1.21 1.22 2.1 2.2 2.3

Organization of Role and Reference Grammar Universal oppositions underlying clause structure Components of the layered structure of the clause The layered structure of the clause in English Layered structure of the clause with constituent and operator projections English clause structure English copular construction Japanese clause structure Structure of Turkish example in (1.7b) English syntactic templates (simplified) from the syntactic inventory Combining syntactic templates from the syntactic inventory Structure of (1.14a) Structure of (1.14b) English translation of (1.14b) The structure of (1.18a ) The structure of (1.19a) English non-predicative PP English predicative PP Yagua predicative PP ‘inside the canoe’ The layered structure of the NP (preliminary) The layered structure of the NP in English The layered structure of the NP in English with NP-initial position Demonstratives in English and Lakhota The structure of the Dyirbal sentence in (1.22b) The structure of the English translation of (1.22b) Semantic representation of Has the tall man been crying? Continuum from verb-specific semantic roles to grammatical relations Thematic relations continuum in terms of logical structure argument positions

page 2 4 4 7 12 14 14 14 14 15 15 17 17 17 22 22 23 23 23 25 25 27 27 29 30 53 54 58 ix

List of figures

x

2.4 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 4.1 4.2 4.3 4.4 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11 5.12 5.13 5.14 5.15 5.16 5.17 5.18

Actor–undergoer hierarchy (preliminary) Coding of referents in terms of possible functions Predicate focus in English Narrow focus in English Two types of narrow focus in a single clause, as in (3.9) Predicate focus and sentence focus in Mandarin Clause structure with constituent, operator and focus structure projections Predicate focus vs narrow focus in the interpretation of quantifiers Italian focus structure and the scope of quantifiers Types of privileged syntactic arguments (preliminary) Types of privileged syntactic arguments (revised) Types of privileged syntactic arguments (final) Actor-undergoer hierarchy (revised) Summary of RRG linking system Syntactic templates for a simple clause and an NP in English Organization of Role and Reference Grammar (revised) Organization of Role and Reference Grammar (final) Output of step 1 in (5.5) Output of step 2 in (5.5) Output of step 3 in (5.5) Output of step 4 in (5.5) Result of all steps in (5.5) Abbreviated linking diagram summarizing Figures 5.5–5.9 Linking in sentence with marked undergoer choice with present Passive voice linking of sentence in Figure 5.11 Linking in a WH-question in English Linking in WH-question with preposition stranding in English Linking from semantics to syntax in (5.6b) from German Linking from semantics to syntax in (5.6c) from German Linking from semantics to syntax in (5.7b) from Dyirbal Linking from semantics to syntax in Acehnese possessor raising construction

61 73 77 78 78 79 80 82 84 102 102 105 126 129 131 131 134 137 137 138 138 139 140 140 141 141 142 144 144 145 146

List of figures

5.19 Linking from semantics to syntax in (5.9a) from Lakhota 5.20 Linking from semantics to syntax in (5.9b) from Lakhota 5.21 Interaction of constructional schemas in the semantics-to-syntax linking in (5.4b ) 5.22 Linking from syntax to semantics in simple English transitive clause 5.23 Linking from syntax to semantics in Enga sentence in (5.11) 5.24 Linking from syntax to semantics in English passive construction 5.25 Linking from syntax to semantics in Dyirbal sentence in (5.7a) 5.26 Linking from syntax to semantics in English WH-question 5.27 Linking from syntax to semantics in English WH-question with stranded preposition 5.28 Linking from syntax to semantics in (5.9b) from Lakhota 5.29 Interaction of constructional schemas in the syntax to semantics linking in (5.4a ) 5.30 Interaction of linking with full clause structure 5.31 Syntactic templates for English WH-question and locative inversion constructions 5.32 A simple example of discourse representation structures 5.33 Derivation of predicate focus 5.34 Derivation of completive and contrastive narrow focus 5.35 Derivation of predicate focus in the linking in Figure 5.10 5.36 Discourse representation structures for (5.33a–c) 5.37 Linking directly from discourse representation structure to logical structure to satisfy the completeness constraint in Mandarin 5.38 Linking from semantics to syntax in ne-cliticization 5.39 Focus structure contrasts in Lakhota WH- and yes–no questions 5.40 The pervasive role of discourse-pragmatics in grammar 6.1 Nexus types 6.2 Core coordination in French 6.3 French nuclear juncture 6.4 Sentential coordination in English

147 148 148 151 152 153 154 155 156 156 157 170 171 172 172 172 173 174

175 178 180 182 188 189 191 192

xi

xii

List of figures

6.5 Simplified structure of sentential subordination in (6.18) 6.6 Ad-core subordination in English 6.7 Ad-clausal subordination in English 6.8 Structure of (6.24a) with ad-core and ad-clausal subordinate clauses 6.9 Ad-nuclear subordination in Lakhota 6.10 English sentence containing clausal, core and nuclear junctures 6.11 Asymmetrical vs symmetrical linkage 6.12 English object complement that-clause 6.13 Kewa clausal coordination 6.14 Tonkawa clausal cosubordination 6.15 English core cosubordination 6.16 English core coordination 6.17 Turkish core cosubordination 6.18 Turkish core coordination 6.19 Barai nuclear cosubordination 6.20 Barai nuclear coordination 6.21 Interclausal semantic relations hierarchy 6.22 Interclausal relations hierarchy 6.23 Analysis of English persuade complex sentence forms 6.24 Potential focus domain in clausal (daughter) subordination 6.25 Potential focus domain in ad-core subordination 6.26 The structure of Lakhota complex sentences in (6.58) 6.27 Noun complement that-clause 6.28 English restrictive relative clause 6.29 Structure of English non-restrictive relative clause 6.30 Template for clausal (daughter) subordination 6.31 Templates for non-subordinate nexus types 7.1 Constituent projection for (7.7) 7.2 The syntactic representation of (7.10c) 7.3 Semantics to syntax linking in the French nuclear juncture causative in (7.18a) 7.4 Semantics to syntax linking in the French nuclear juncture causative in (7.18b) 7.5 Linking from semantics to syntax in English control construction in (7.23a) 7.6 Failed linking from semantics to syntax in (7.35a) 7.7 Linking from semantics to syntax in (7.35b) 7.8 Failed linking from semantics to syntax in (7.35c) 7.9 Linking from syntax to semantics in Dyirbal control construction in (7.24b)

193 194 195 195 197 199 199 200 203 203 203 203 204 204 204 204 208 209 211 214 216 220 221 222 222 223 224 230 232 236 237 246 247 247 248 249

List of figures

7.10 Linking from semantics to syntax in English matrix-coding construction in (7.37a) 7.11 Linking from syntax to semantics in English matrix-coding construction 7.12 Linking from semantics to syntax in the Icelandic construction in (7.42b) 7.13 Linking from syntax to semantics in English matrix-coding construction 7.14 Linking from semantics to syntax in Malagasy relative clause in (7.49a) 7.15 Linking from syntax to semantics in Malagasy relative clause in (7.49a ) 7.16 Linking from semantics to syntax in English relative clause in (7.53a) 7.17 Linking from syntax to semantics in English relative clause in (7.53a) 7.18 Linking from semantics to syntax in (7.63a) 7.19 Linking from semantics to syntax in (7.63c) 7.20 The structure of Lakhota complex sentences in (7.66c) and (7.69)

253 253 256 257 262 264 266 266 272 272 276

xiii

Tables

1.1 Semantic units underlying the syntactic units of the layered structure of the clause 1.2 Operators in the layered structure of the clause 1.3 Operators in the layered structure of the NP 2.1 Tests for Aktionsart classes 2.2 Tests for determining predicate classes 2.3 Lexical representations for Aktionsart classes 2.4 Definitions of thematic relations in terms of logical structure argument positions 2.5 Macrorole number and transitivity 3.1 Quantifier hierarchy from Ioup (1975) and Kuno et al. (1999) 4.1 Patterns of restricted neutralizations of semantic roles 4.2 PSA defaults in accusative and ergative systems 5.1 Constructional schema for English passive (plain) 5.2 Constructional schema for Sama antipassive 5.3 Constructional schema for English WH-questions 5.4 Constructional schema for Sama WH-question formation 5.5 Constructional schema for Sama reflexivization 5.6 Constructional schema for English reflexivization 6.1 Juncture and nexus in the NP 7.1 Constructional schema for English ‘conjunction reduction’ 7.2 Constructional schema for English ‘VP’ ellipsis 7.3 Constructional schema for French nuclear juncture causative construction in (7.18) 7.4 Constructional schema for English resultative construction in (7.19a) 7.5 Constructional schema for English control constructions in (7.23) 7.6 Constructional schemas for English matrix-coding constructions 7.7 Constructional schema for Malagasy relative clause constructions xiv

page 5 9 24 35 39 45 55 64 86 99 100 132 133 133 134 135 169 221 231 235 238 239 250 258 265

List of tables

7.8 Constructional schema for English relative clause constructions 7.9 Potential focus domain in English complex sentence constructions 7.10 Discourse functions of ‘extracted’ element in ‘extraction’ constructions 7.11 Markedness of narrow focus

267 278 285 291

xv

Preface

This book is an introduction to Role and Reference Grammar, which updates the presentation of the theory presented in Syntax: structure, meaning and function. It includes the major developments in the theory since the publication of Syntax in 1997. It can be used as a textbook for teaching Role and Reference Grammar, and while it does not contain any exercises, the exercises from Syntax can be used with this book as well. The following table relates the exercises from the earlier book to the chapter topics in this one. Exercises in Syntax

Chapter in this book

Chapter 2 Chapters 3, 4 Chapter 5 Chapter 6 Chapter 7 Chapter 8 Chapter 9

Chapter 1 Chapter 2 Chapter 3 Chapter 4 Chapter 5 Chapter 6 Chapter 7

These exercises (with password-protected solutions) are available online www.cambridge.org/052101056X. This book has benefited greatly from comments and suggestions from many people over the course of its long gestation. I would especially like to thank the many participants in courses on Role and Reference Grammar, both at the University at Buffalo and at universities around the world, for their indispensable insights, comments and suggestions, especially Jeruen Dery, Klaus Michael Geyer, Anja Latrouite, Heike Lehnert, Mark Miller, John Roberts and Janick Wrona. I would also like to thank Delia Bentley, Balthasar Bickel, Daniel Everett, Rolf Kailuweit, Craig Kopris, Kwee Tjoe Liong, Beth Levin, Laura Michaelis, Toshio Ohori, Mitsuaki Shimojo, Richard Weist and David Wilkins for their valuable input.

xvii

Abbreviations

A, ACT AT ABS ACC ACS ACV ADV AFD ANTI AOR APPL ART ASP ATV AUG AUH AUX BNU CAUS CL CLM CONT COP DAT DCA DCT DEC DEF DEIC DEM DEP DET DIR dl DOBJ xviii

actor actor of transitive verb absolutive accusative accessible activated adverb actual focus domain antipassive aorist applicative article aspect active voice stem augment actor–undergoer hierarchy auxiliary brand new – unanchored causative classifier clause-linkage marker continuative copula dative direct core argument direct declarative definite deictic demonstrative dependent determiner directional dual direct object

List of abbreviations

d-S DS DSTR DT DTR ERG EVID EVQ F FOC FPAST FUT GEN HON HPSG HS IF IMP IMPER IMPF IN INA INAN INCL IND INF INFER INGR INST INT INTR IRR ITER IU LDP LFG LNK LOC LS LSC LSNP M MOD MR

derived intransitive subject different subject distributive different topic detransitivizer ergative evidential event quantification feminine focus far past future genitive honorific Head-Driven Phrase Structure Grammar hearsay illocutionary force imperative impersonal imperfective inchoative inactive inanimate inclusive indicative infinitive inferential ingressive instrumental interrogative intransitive irrealis iterative information unit left detached position Lexical-Functional Grammar linker locative logical structure layered structure of the clause layered structure of the noun phrase masculine modality macrorole

xix

xx

N NASP NEG NEGPAST NM NMR NOM NONSIM NP NPFP NPIP NPP ns NUC NUM OBJ OBL OBLG P PART PASS PAST PER PERF PFD pl PM PoCS POL PP PrCS PRDM PRED PRES PRESUM PRFV PRO PROC PROG PROP PROX PRT PSA PSBL


neuter nominal aspect negative negative past noun marker non-macrorole nominative non-simultaneous noun phrase NP-final position NP-initial position non-past progressive non-singular nucleus number object oblique obligation preposition/postposition participle passive past tense periphery perfect potential focus domain plural proper noun marker postcore slot politeness marker prepositional/postpositional phrase precore slot predicate marker predicate present presumptive evidential perfective pronoun process progressive proper proximate particle privileged syntactic argument possible


PSTP Q QLT QNT RDP REFL REL REMPST RES RRG S SEML SEQ sg SH SIM SP SPEC SS STA SUBJ SUFF TM TNS TOP TPAST TR U, UND UT V VBLZ VIS VP XP YPAST

past participle question marker quality quantifier right-detached position reflexive relative remote past result Role and Reference Grammar intransitive subject semelfactive sequential singular subject honorific simultaneous same privileged syntactic argument specific same subject status subject suffix terminal marker tense topic today’s past transitive undergoer undergoer of transitive verb verb verbalizer visual evidence evidential verb phrase phrase of category X yesterday’s past

xxi

Introduction

This book is concerned with how structure, meaning and communicative function interact in human languages. Language is a system of communicative social action in which grammatical structures are employed to express meaning in context. While all languages can achieve the same basic communicative ends, different languages use different linguistic means to achieve them, and an important aspect of these differences concerns the divergent ways syntax, semantics and pragmatics interact across languages. For example, the noun phrase referring to the entity being talked about (the ‘topic’) may be signalled by its position in the clause, by its grammatical function, by its morphological case, or by a particle in different languages, and moreover in some languages this marking may have important grammatical consequences and in other languages it may not. The framework in which this investigation is to be carried out is Role and Reference Grammar [RRG].1 RRG grew out of an attempt to answer two basic questions, which were originally posed during the mid-1970s: (1) what would linguistic theory look like if it were based on the analysis of languages with diverse structures such as Lakhota, Tagalog and Dyirbal, rather than on the analysis of English?, and (2) how can the interaction of syntax, semantics and pragmatics in different grammatical systems best be captured and explained? Accordingly, RRG has developed typologically motivated descriptive tools and theoretical principles which are designed to expose this interaction and offer explanations for it. It posits three main representations: (1) a representation of the syntactic structure of sentences, which corresponds closely to the actual structural form of utterances, (2) a semantic representation representing important facets of the meaning of linguistic expressions, and (3) a representation of the information (focus) structure of the utterance, which is related to its communicative function. There is a set of rules, called the linking algorithm, which relates the syntactic and semantic representations to each other, and discourse-pragmatics plays a role in the linking. From an RRG perspective, one of the most important ways in which languages differ from each other is in terms of the manner in which discourse-pragmatics interacts with the linking between syntax and semantics. This is summarized in Figure 1. 1

A bibliography of work in RRG is available on the RRG web site (linguistics.buffalo.edu/ research/rrg.html), along with papers, master’s theses and dissertations in PDF format for downloading. It is the best source for the most up-to-date work in the theory.

1

2

introduction

Linking Algorithm

SEMANTIC REPRESENTATION

DISCOURSE-PRAGMATICS

SYNTACTIC REPRESENTATION

Figure 1 Organization of Role and Reference Grammar

RRG seeks to be more than a descriptive framework for the analysis of languages; it also strives to provide an explanatory framework for the analysis of language acquisition and language processing. See Van Valin and LaPolla (1997), ‘Epilog’, and the references cited therein, also Van Valin (1998, 2001a, 2002, 2005), Butler (2004), Weist (2002), and Weist et al. (2004). Thus, this exploration of the syntax, semantics and pragmatics interface will also be an explication of RRG; this presentation presupposes no prior acquaintance with the theory.2 The discussion will proceed as follows. The first five chapters deal with simple clauses: in chapter 1 the conception of clause structure proposed in RRG will be detailed; chapter 2 brings in the theory of lexical representation and semantic roles; in chapter 3 the notion of information structure is introduced and related to clause structure; chapter 4 introduces grammatical functions and their interaction with semantic roles and information structure; and chapter 5 presents the algorithm which links syntactic and semantic representations in simple clauses. The next two chapters deal with complex sentences: in chapter 6 the approach to the syntax and semantics of clause linkage is outlined, and in chapter 7 the linking algorithm is expanded to deal with argument linking in complex sentences. 2

It does presuppose a basic introduction to syntax, as found in, e.g., Carnie (2002), Tallerman (1998), Van Valin (2001b). See Butler (2003) for a critical introduction to RRG along with a comparison of it with Functional Grammar and Systemic Functional Grammar.

1

Syntactic structure

1.0

General considerations

The first step in the exploration of the syntax, semantics and pragmatics interface in the grammatical systems of human languages is to characterize the nature of syntactic structure, which includes the structure of clauses, of adpositional phrases and noun phrases. There are two fundamental aspects of clause structure which every theory must deal with; they may be termed relational and non-relational . Relational structure is concerned with relations between a predicate and its argument(s); they may be syntactic, semantic, pragmatic or some combination thereof; these relations will be the topics of chapters 2–4. Non-relational structure is concerned with the hierarchical organization of phrases, clauses and sentences; it is the topic of this chapter. From an RRG point of view, there are two general considerations that a theory of clause structure must meet. They are given in (1.1). (1.1) a.

b.

General considerations for a theory of clause structure: A theory of clause structure should capture all of the universal features of clauses without imposing features on languages in which there is no evidence for them. A theory should represent comparable structures in different languages in comparable ways.

These are very strong conditions, especially considering that RRG does not posit any abstract underlying syntactic representations; the syntactic representation of a sentence corresponds closely to its actually occurring form.

1.1

The layered structure of the clause

These considerations lead to a very different conception of clause structure in RRG from that assumed in other approaches. First, because it is concerned with the interplay of syntax, semantics and pragmatics in grammatical systems, the representation of clauses must allow for the representation of all of these factors, where necessary. This renders a representation in terms of purely syntactic features highly inappropriate. Second, the theory is greatly concerned 3

4

syntactic structure

with typological issues. In particular, it seeks to uncover those aspects of clause structure which are found in all human languages; hence the conception of clause structure it posits must be equally applicable to free-word-order, flat-syntax languages such as Dyirbal and Malayalam, to head-marking languages like Lakhota and Tzotzil (see section 1.4), and to fixed-order, configurational, dependentmarking languages like English and Icelandic. It must, further, be applicable to these languages without recourse to positing abstract underlying representations and derivations from abstract to overt representations. The RRG notion of (non-relational) clause structure is called ‘the layered structure of the clause’ and it is based on two fundamental contrasts: between the predicate and non-predicating elements, on the one hand, and, among the non-predicating elements, between arguments and non-arguments, on the other, i.e. between those NPs and adpositional phrases which are arguments of the predicate and those which are not. These contrasts are found in all languages, regardless of whether they are configurational or non-configurational, head-marking or dependent-marking, free-word-order or fixed-word-order. On this view, the primary constituent units of the clause are the ‘nucleus’, which contains the predicate (usually a verb), the ‘core’, which contains the nucleus and the arguments of the predicate, and a ‘periphery’, which subsumes non-arguments of the predicate, e.g. setting locative and temporal phrases (see section 1.5). This may be informally represented as in the box diagrams in Figures 1.1 and 1.2. The semantic motivation for these units is summarized in Table 1.1. The semantic representations which underlie the notion of core argument will be presented in chapter 2.

Predicate

+ Arguments

Non-Arguments

Figure 1.1 Universal oppositions underlying clause structure

CLAUSE CORE PERIPHERY

NUCLEUS

CLAUSE CORE

PERIPHERY

ARGUMENTS

ADJUNCTS

Dana

saw

Pat

yesterday in the library

NUCLEUS Figure 1.2 Components of the layered structure of the clause

1.1 The layered structure of the clause

Table 1.1 Semantic units underlying the syntactic units of the layered structure of the clause Semantic element(s)

Syntactic unit

Predicate Argument in semantic representation of predicate Non-arguments Predicate + Arguments Predicate + Arguments + Non-arguments

Nucleus Core argument Periphery Core Clause (= Core + Periphery)

Since these hierarchical units are defined semantically and not syntactically, they are not dependent upon either immediate dominance or linear precedence relations. Accordingly, the elements in these units may in principle occur in any order, if a given language permits it. This is illustrated in (1.2) from Dyirbal, an Australian Aboriginal language with exceptionally free word order; all of the sentences in (1.2) mean ‘The man speared the wallaby in the mountains.’ (1.2) a.

b. c. d.

Dyirbal (Dixon 1972) ya a-ŋguCORE d¸urga-uNUC Bayi barganCORE baŋgul DET.ABS wallaby-ABS DET.ERG man-ERG spear-TNS mountains-LOC gambi- aPER. Baŋgul ya aŋguCORE gambi aPER bayi barganCORE d¸urgauNUC. man mountains wallaby speared DyurgauNUC gambi aPER bayi barganCORE baŋgul ya aŋguCORE. speared mountains wallaby man Bayi barganCORE gambi aPER d¸urgauNUC baŋgul ya aŋguCORE. wallaby mountains speared man ‘The man speared the wallaby in the mountains.’

Bayi bargan ‘wallaby (ABS)’ and baŋgul yaaŋgu ‘man (ERG)’ are arguments in the core of this clause (i.e. ‘core arguments’), regardless of the word order in the clause, because they are arguments of the predicate d¸urga- ‘spear’, which is the nucleus of the clause. Gambia ‘mountains (LOC)’ is not an argument of d¸urga-, and therefore it is not part of the core; it is, rather, an element in the periphery of the clause. The hierarchical structure of the clause is semantically motivated and not strictly syntactically based. There are additional elements which may occur in a simple sentence, i.e. a single-clause sentence. The first is the ‘precore slot’ [PrCS], the position in which question words appear in languages in which they do not occur in situ, e.g. English, Italian, Zapotec; it is also the location in which the fronted element in a sentence like B e a n s o u p I can’t stand appears. This position is clause-internal but coreexternal. There is also a ‘postcore slot’ [PoCS] in some verb-final languages, e.g.

5

6

syntactic structure

Japanese (Shimojo 1995), Dhivehi (Indo-Aryan; Cain and Gair 2000). In Dhivehi, both WH-words and non-WH NPs can occur in the postcore slot; they are in small caps. (1.3) a.

a .

b.

b .

Al¯ı bun¯ı k¯ı ke ta? Ali say.PAST.FOC what Q ‘What did Ali say?’ Al¯ı k¯ıke bun¯ı ta? Ali what say.PAST.FOC Q ‘What did Ali say?’ M¯ale ul.un¯ıma aharen bon¯ı ais Male be.PAST.PROG.when 1sg drink.PRES.FOC ice ‘When in Male, it is ice cream that I eat.’ M¯ale ul.un¯ıma ais kur¯ımu bon¯ı Male be.PAST.PROG.when ice cream drink.PRES.FOC ‘When in Male, it is I who eat ice cream.’

kur¯ı m u. cream ah aren. 1sg

In addition to a clause, a simple sentence may also include a phrase in a detached position, most commonly in the ‘left-detached position’ [LDP]. This is the location of sentence-initial elements, most commonly adverbials, which are set off from the clause by a pause, e.g. Ye s t e r d ay , I bought myself a new car or A s f o r J o h n , I haven’t seen him in a couple of weeks. There is also a ‘rightdetached position’ [RDP], as in sentences like I know them, t h o s e b o y s . When the element in a detached position functions as a semantic argument of the verb, there is normally a resumptive pronoun in the core referring to it. A language which clearly indicates the precore slot – left-detached position contrast is Tzotzil (Aissen 1987, 1992), a VOS Mayan language. In this language the contrast is marked both intonationally and morphologically. This is exemplified in (1.4). (1.4) a.

b.

c.

d.

e.

’

ʔ i-Ø-s-pet

’

lokel ʔ antz ti t ul-e. ASP-3ABS-3ERG-carry away woman DEF rabbit-DEF ‘The rabbit carried away the woman.’ ’ Buchu Ø-s-tam? who 3ABS-3ERG-take ‘Who took it?’ ’

’

Kusi ch-Ø-a-kan? what ASP-3ABS-2ERG-want ‘What do you want?’ ʔ un, voʔ on la ch-i-kom ʔ un. Voʔ ot la ch-a-bat 2sg CL ASP-2ABS-go CL, 1sg CL ASP-1ABS-stay CL ‘It’s you who’s going, I’m staying.’ ʔa ti tzeb-e, ʔ i-Ø-s-saʔ s-malal. TOP DEF girl-DEF ASP-3ABS-3ERG-search 3ERG-husband ‘The girl, she searched for her husband.’

1.1 The layered structure of the clause

Question words occur in immediately preverbal position, as in (b) and (c); there is no pause following them. An NP may also occur in this position, as in (d). This immediately preverbal position is the precore slot. A left-detached position phrase is both set off by a pause and marked by the ‘topic particle’ʔa, as in (e). The contrast in the discourse functions of phrases in the two positions is illustrated by the following text excerpt (Aissen 1987:158). (1.5)

a.

b.

[Something had landed at the foot of the tree, they went to look. There was a straw mat. ‘Hell, what could it be? Come on, let’s untie the straw mat!’ the two men said to each other. They untied it. You know what?–] ʔ un. Tzeb san-andrex la te Ø-s-ta-ik girl San Andreas CL there 3ABS-3ERG-find-pl CL ‘A San Andreas girl they found there.’ ʔa

ti tzeb san-antrex ʔ un-e, ʔ i-Ø-y-ik-ik la TOP DEF girl San Andreas CL-DEF ASP-3ABS-3ERG-take-pl CL ech’el ʔ un. away CL ‘The San Andreas girl, they took her with them.’

In (1.5a) tzeb san-andrex ‘a San Andreas girl’ is a piece of unexpected new information and is the major focus of the assertion, and consequently it must occur clause-internally in order to be within the domain of the illocutionary force operator. It is preverbal but is not set off by a pause and is not marked by the topic marker ʔa, and accordingly it is in the precore slot. In (b), on the other hand, the same NP is marked by ʔa and set off by a pause, and, in addition, it has already been introduced as a significant participant in the context. Therefore here it is in the left-detached position. An English sentence containing a precore slot is presented in Figure 1.3. In this representation, an arrow indicates that the periphery is an optional modifier of the core. The periphery will be discussed further in section 1.5. In the sentence in Figure 1.3, Robin and Pat are arguments within the core, but Robin is a direct core argument, while Pat is an oblique core argument, because it is marked by a preposition. In languages with case systems, nominative, accusative and dative (or ergative, absolutive and dative) are the direct cases, while the other cases count as oblique. SENTENCE CLAUSE PrCS

CORE NP

NUC

PERIPHERY PP

PRED NP What did Robin

V show to Pat

PP

ADV

in the library yesterday?

Figure 1.3 The layered structure of the clause in English

7

8

syntactic structure

The RRG conception of the layered structure of the clause is thus a semantically based theory of non-relational syntactic structure; that is, the fundamental units in the hierarchical organization of sentences and clauses are semantically motivated by the contrast between predicate and argument, on the one hand, and that between XPs, i.e. NPs and PPs, which are related to the predicate and those which are not, on the other (see Table 1.1). These units are, however, syntactic units. There are dissociations between the semantic motivations and the syntactic instantiation of these concepts. For example, while the notion of nucleus is based on the semantic notion of predicate, in some languages a nucleus may contain more than the predicate, e.g. an incorporated argument. In Lakhota, the noun cˇ há¸ ‘tree, wood’ can be compounded with the verb kaksá ‘chop’ to create cˇ ha¸ káksa ‘wood-chop’, a noun + verb combination that can function as the nucleus of a Lakhota clause. The notion of core argument is motivated by the notion of argument in the semantic representation of the verb, but there are clear instances in English and other languages in which this correlation does not hold. For example, the dummy it in it is snowing occupies a core argument position but is not a semantic argument of snow. Furthermore, in a passive construction like The bagel was eaten by Chris, Chris is a semantic argument of eat but occurs as an adjunct in the periphery. Hence while the semantic distinctions in Table 1.1 motivate the syntactic distinctions, there is no absolute correlation between them. There is an interesting difference between the universal and non-universal aspects of clause structure. The universal aspects (the nucleus, core, periphery and clause) are all semantically motivated, as shown in Table 1.1. The non-universal aspects (the detached phrases, the extra-core slots) are not semantically motivated; rather, they seem to be pragmatically motivated (or at least are associated with constructions that have strong pragmatic conditions on their occurrence). A major difference between RRG and some other theories is that the category of VP, which plays a central role in Chomskyan approaches, has no analogue in the layered structure of the clause.1

1.2

Operators

In Figure 1.3, the auxiliary verb did is not attached to anything, and this is because it is not part of the nucleus, core or periphery. It is, rather, the morphological realization of a tense ‘operator’ which modifies the clause. Grammatical categories like aspect, tense and modality are treated as operators modifying different layers of the clause. Each of the clause levels may be modified by one or more operators. The operators are summarized in Table 1.2. The nuclear operators have scope over the nucleus; they modify the action, event or state itself 1

The source of VPs in languages which have them will be discussed in section 3.5.

1.2 Operators

Table 1.2 Operators in the layered structure of the clause Nuclear operators: Aspect Negation Directionals (only those modifying orientation of action or event without reference to participants) Core operators: Directionals (only those expressing the orientation or motion of one participant with reference to another participant or to the speaker) Event quantification Modality (root modals, e.g. ability, permission, obligation) Internal (narrow scope) negation Clausal operators: Status (epistemic modals, external negation) Tense Evidentials Illocutionary force

without reference to the participants. Core operators modify the relation between a core argument, normally the actor, and the action; this is especially true of core directionals and modality. Clausal operators, as the name implies, modify the clause as a whole. They fall into two groups, one containing tense and status, and the other evidentials and illocutionary force. Tense and status situate the proposition expressed by the clause within temporal and realis–irrealis continua; evidentials indicate the epistemological basis of the state of affairs (the proposition plus tense and status operators) expressed, i.e. how the speaker came to have the information being uttered, while illocutionary force specifies the type of speech act. Hence evidentials and illocutionary force are modifiers of the sentence or utterance as a whole, rather than one of its constituent clauses; they are thus ‘sentential’ in nature. They occur only in main (root) clauses, i.e. clauses which are immediately dominated by the sentence node. Negation is the only operator that occurs at all three levels: nuclear negation has only the nucleus in its scope, core negation has one or more core arguments (and possibly also the nucleus) in its scope, and clausal negation has the entire clause in its scope. The classification of a particular operator as nuclear, core or clausal is a direct function of its meaning. No language need have all of these operators as grammatical categories; for example, English, unlike Kewa and Quechua, does not have evidentials as a grammatical category. The only operators which every language has are illocutionary force and negation. Examples of operators from a variety of languages are given in (1.6)–(1.12). The Kewa examples in (1.6) involve aspect, nuclear directionals (nuclear because they modify the motion of the action, not a participant), and tense. The Turkish examples in (1.7) illustrate aspect and tense in (a) and modality, status and tense

9

10

syntactic structure

in (b). Both Kewa and Turkish are verb-final, left-branching languages, and consequently the operator-bearing morphemes follow the verb; this is also the case in Barasano in (1.10) and Amele in (1.12). English and Tiwi, which are verb-medial, right-branching languages, exhibit the opposite arrangement: the operator-bearing morphemes precede the verb, as in (1.8) and (1.11). Note that the order of tense, aspect and the verb in the Tiwi example in (1.11) is the mirror image of the order in the Kewa, Turkish and Barasano examples. (1.6) a.

b.

(1.7) a.

b.

(1.8) a. b. c.

Kewa (Papua-New Guinea; Franklin 1971) Íra-pa-niaa-ru. cook-PERF-down-1sgPAST ‘I burned it downward (as a hill).’ Íra-pa-saa-ru. cook-PERF-up-1sgPAST ‘I burned it upward (as a hill).’ Turkish (Watters 1993) Gel-ıyor-du-m. come-PROG-PAST-1sg ‘I was coming.’ Gel-emi-yebil-ir-im. come-ABLE.NEG-PSBL-AOR-1sg ‘I may be unable to come.’ English He may be leaving soon. She was able to see them. Will they have to be leaving?

(V-ASP-DIR-TNS)

(V-ASP-DIR-TNS)

(V-ASP-TNS)

(V-MOD-STA-TNS)

(IF/TNS-STA-ASP-V)2 (IF/TNS-MOD-V) (IF/TNS-MOD-ASP-V)

(1.9)

Korean (Yang 1994) An-tul-li-wu-(si)-lswuiss-cianh-ass-keyss-up-nita. NEG-hear-CAUS-PASS-(SH)-ABLE-NEG-PAST-PRESUM-POL-DEC (NEG-V-MOD-NEG-TNS-EVID-IF) ‘(I) guess that (he) [HON] might not be heard.’

(1.10)

Barasano (Tucano, South America; Jones and Jones 1991) Gahe-r-˜ib-˜i bota-ri k˜ea˜ -kudi-ka-bã ˜ıdã. other-day post-pl chop-ITER-FPAST-VIS 3pl (V-ASP-TNS-EVID) ‘The next day they went from place to place chopping down posts.’

(1.11)

Tiwi (Australia; Osborne 1974) ŋə -ru-untiŋ-apa. ¯¯ 1sg-PAST-PROG-eat ‘I was eating.’

2

(TNS-ASP-V)

The reason that illocutionary force and tense are linked in the English examples is that, in English, illocutionary force is indicated by the position of the tense marker in the main clause: interrogative by core-initial tense, declarative by core-internal tense, and imperative by no tense.

1.3 Formal representation: clauses

(1.12)

Amele (Papua-New Guinea; J. Roberts 1987, 2002) Age bel-ad-ein. 3pl go-DSTR-3plREMPST ‘They went in all directions.’

(V-EVQ-TNS)

The Amele example illustrates event quantification; the morpheme -ad- signals that there are multiple actions of the verb; if it were omitted, yielding age bel-ein (3pl go-3plREMPST) ‘they went’, the sentence would mean that there was only one event of going. Event quantification is a core operator, because the core is the minimal expression of an event or state of affairs. One of the major claims regarding operators made in RRG is that the ordering of the morphemes expressing operators with respect to the verb indicates their relative scopes. That is, taking the nucleus as the reference point, the morphemes realizing nuclear operators should be closer to the nucleus than those expressing core operators, and those manifesting clausal operators should be outside of those signalling nuclear and core operators. The ordering of the operators in the sentences in (1.6)–(1.12) exemplifies this pattern. In every example involving aspect and tense, the aspect marker occurs closer to the verb than the tense marker. Modality and status markers occur outside of aspect (if both occur) but inside of tense, as in (1.7b), (1.8), and (1.9). Evidentials and illocutionary force, the two sentential clausal operators, occur outside of tense in (1.9) and (1.10). This claim assumes, crucially, that a relative order among the morphemes with reference to the nucleus can be established. If, for example, tense is a prefix and aspect is a suffix on the verb, then no relative ordering can be determined and therefore this claim is not applicable; if, on the other hand, both are suffixes, then the claim is that the aspect suffix would be between the verb stem and the tense suffix. This is exemplified in all of the examples in (1.6)–(1.12). A large number of languages are surveyed in Foley and Van Valin (1984) and Bybee (1985), and no exceptions to this general claim are found. Thus, the ordering of operator-expressing affixes in, e.g., Turkish follows exactly the same principle as auxiliary elements in English. It should be noted that within a group of operators at the same level, there is some variation in ordering. Aspect is normally inside of nuclear directionals, but in a few languages it occurs outside of them. Tense and status vary in their position relative to each other across languages, but they are always inside of evidentials and illocutionary force; universally, illocutionary force is always the outermost operator over the clause.

1.3

Formal representation of clause structure

Since operators are technically not part of the nucleus, core or periphery, but rather are modifiers of these units and combinations thereof, it is reasonable that they should be represented separately from the predicates and arguments that they modify. Moreover, predicates and arguments are subject to

11

12

syntactic structure

language-specific constraints on their ordering, while the primary principle governing the ordering of operators is the universal scope constraint mentioned above. The main language-specific consideration affecting their occurrence is the basic word-order type of the language, i.e. whether the language is right-branching or left-branching (Dryer 1992), which governs whether the operators are predominantly prefixes or suffixes, if they are bound morphemes, or whether they occur before or after the nucleus, if they are free morphemes. But the ordering among them is determined by the scope principle. In order to be able to capture the differences between the restrictions on predicates and arguments, on the one hand, and operators, on the other, Johnson (1987) proposed a formalization of the layered structure of the clause in which predicates and their arguments are represented in a distinct projection from the one representing operators. This formalization he termed a ‘projection grammar’. The general schema of a projection grammar representation of the layered structure of the clause is given in Figure 1.4. The top part is called the ‘constituent projection’, the bottom the ‘operator projection’. The two projections are joined through the nucleus, which is the central element in the clause both in terms of defining the range of possible arguments, on the one hand, and being the primary entity to which the operator grammatical categories are oriented, on the other. (This can be seen most clearly in Figure 1.4a.) In the operator projection, the scope of the operator is indicated by the unit which is the SENTENCE SENTENCE CLAUSE

(LDP)

CLAUSE

CORE NUC

(XP) (XP)

(RDP) (PoCS)

CORE

(PrCS) XP

PRED

(XP)

NUC

(XP)

(XP)

PRED

V

V

Aspect Negation Directionals

NUC

Directionals Event quant Modality Negation

CORE

CLAUSE SENTENCE

(a)

NUC

Aspect

NUC

Negation

NUC/CORE

Status Tense Evidentials Illocutionary Force

CORE

Event quantification

CORE

Modality

CORE

(b)

Directionals

Negation (Internal)

CLAUSE

Status

CLAUSE

Tense

CLAUSE

Evidentials

CLAUSE

Illocutionary Force

SENTENCE

Figure 1.4 Layered structure of the clause with constituent and operator projections

1.3 Formal representation: clauses

target of the arrow. Each operator at a given level is so represented, and if there is more than one, e.g. both tense and illocutionary force, then the relative scopes are explicitly indicated. Examples from English, Japanese and Turkish are given in Figures 1.5–1.8 on the next page. The ‘PRED’ node under the nucleus node is required, because, as will be seen in chapter 6, there are constructions containing a nucleus in which the verb does not predicate; rather, it serves as a modifier of another verb. Hence there is a contrast between nuclei which contain a predicating verb and those which contain a non-predicating verb. In an English copular construction, as in Figure 1.6, the nucleus consists of the predicate nominal, adjective or PP, and be is an auxiliary verb serving as a tense-carrier.3 By separating the constituent from the operator projections, the representations in Figures 1.5 through 1.8 show clearly the similarities in operator orders despite the manifest differences in the predicate-argument arrangement in the three languages. In RRG, syntactic representations are not specified by phrase-structure rules or the like; rather, the different patterns are stored as ‘syntactic templates’ in a ‘syntactic inventory’. While the layered structure of the clause is universal, there is substantial cross-linguistic variation with respect to the syntactic templates in the syntactic inventory of each language, e.g. languages which lack a precore slot or postcore slot will not have the templates for them, and in some languages the core templates will be ordered (e.g. English), while in others they will not (e.g. Dyirbal). In languages with fixed word order, this will be specified in the syntactic templates. In languages with flexible word order, the templates would be unordered to various degrees, e.g. the verb position is specified but the argument positions are unordered with respect to each other; additional factors, e.g. information structure (see section 5.4.2), would determine the ordering in any particular utterance. Examples of simplified English templates are given in Figure 1.9 on p. 15; the operator projection and the peripheries have been omitted. Fully specified templates will be presented in chapter 5. In these examples, the templates contain specifications for the NPs and PPs that fill the positions in the core. The template inventory for a language could in fact be simplified if these category labels were omitted; in that case, for example, there would be a single underspecified template replacing the core-5 and core-6 templates for verbs like give in both of their linking possibilities, i.e. NP V NP PP and NP V NP NP. Templates are merged to create the constituent projection of the syntactic representation of a sentence. In the English syntactic inventory, there are templates for each of the various extracore positions (left-detached position, right-detached position, precore slot), and they combine with the core templates to constitute a clause and a sentence. This is illustrated in Figure 1.10 also on p. 15. 3

Be is represented as an auxiliary verb in the nucleus only when it is required for nucleus formation, as with predicate adjectives or nominals, or with passive verbs. That is, be occurs in these constructions even when there are no operators, e.g. in an infinitive like Kim wants to be a doctor. Auxiliaries which function exclusively as operators, e.g. be in the progressive or have in the perfect, are represented in the operator projection only.

13

14

syntactic structure SENTENCE

SENTENCE

CLAUSE

CLAUSE

CORE

CORE NP

NUC

NP

NUC

PRED

AUX

PRED

PP

V

Pat was in the library.

Will they have to be leaving?

PP

V ASP

NUC

MOD TNS IF

NUC CORE

CORE

TNS CLAUSE

CLAUSE

IF

CLAUSE SENTENCE

Figure 1.6 English copular construction

Figure 1.5 English clause structure

SENTENCE

SENTENCE

CLAUSE

CLAUSE

CORE

CORE

PERIPHERY

NP

NP

NUC

NUC

PRED

PRED

Taroo ga Kazue no uti de hon o

Gel- emi-

yon-da ka?

yebil- ir-

come-ABLE.NEG-PSBL-AOR-1sg

GEN house in book ACC read-PAST-Q

im Ø DEC

V

V

‘Did Taroo read a/the book at Kazue's house?’

PRO

V

V

PP NOM

CLAUSE SENTENCE

NUC

NUC

CORE< MOD

CORE

CORE
CORE NUC
NUC > CLAUSE

IF

> CLAUSE SENTENCE

Figure 1.13 The structure of (1.18a ) SENTENCE CLAUSE
CLAUSE IF

>CLAUSE SENTENCE

Figure 1.14 The structure of (1.19a)

semantic information to the clause and do not license the argument they mark. Rather, their argument is licensed by the predicate, i.e., it is a core argument; these prepositions are a function of the semantics of the predicate and are in effect free-morphemic case markers assigned by it. Thus, in a sentence like the one in Figure 1.3, to Pat is a non-predicative PP functioning as a core argument, while in the library is a predicative PP functioning as an adjunct. There is a third function that a PP can have: it can mark an argument of the verb but nevertheless contribute its semantics to the clause. The best example of this is the PP with a verb like put: such verbs require a locative expression,

1.6 Adpositional and noun phrase structure

but the choice of locative preposition is not determined by the verb, as (1.20) shows. (1.20)

Kim put the book in / on / next to / behind / on top of / under the box.

In this sentence, the different prepositions contribute an important component of the meaning of the sentence, unlike the preposition to with a verb like give or show, and therefore they must be considered predicative prepositions. PPs like those in (1.20) will be referred to as ‘argument-adjuncts’ in the core. It will be shown in the discussion of reflexivization in section 5.3 that argument-adjuncts behave differently from arguments in crucial ways. The two types of adposition would be symbolized differently in projection grammar representations of the layered structure of the clause. Since nonpredicative adpositions mark core arguments, the phrases in which they occur would be PPs in the core; the adposition would be treated the same as a case marker, and in particular it would not be considered the head of the phrase. Predicative adpositions, on the other hand, are the head and nucleus of the phrase in which they occur, and their object is in fact their single argument. The internal structures of an argument (non-predicative) PP and a predicative PP are given in Figure 1.15. Figure 1.15b presents the typical structure of a predicative adpositional phrase in a dependent-marking language. Argument-adjuncts have the internal structure of a predicative PP (Figure 1.15b) but occur in the core. PP

PP

CORE P NP

P

NUC P

NP

PRED P to

Pat

Figure 1.15a English non-predicative PP

in

the library

Figure 1.15b English predicative PP

The structure is different in a head-marking language like Yagua (Payne and Payne 1989), in which the adpositional head bears a morpheme expressing its argument (object), e.g. rá-viimú ju¸ muñu´ (3INAN-inside canoe) ‘inside the canoe’ [lit. ‘inside-iti canoei ’]. As with the other head-marking constructions, the dependent can be omitted, with the result being a grammatical phrase of the same type, e.g. rá-viimú ‘inside it’. The structure of this example is given in Figure 1.16. PP CORE P

(NP)

PRO NUC P PRED P rá-viimú (jumuñú)

Figure 1.16 Yagua predicative PP ‘inside the canoe’

23

24

syntactic structure

Table 1.3 Operators in the layered structure of the NP NuclearN operator: Nominal aspect (count–mass distinction, classifiers in classifier languages) CoreN operators: Number Quantification (quantifiers) Negation NP operators: Definiteness Deixis

It has long been noted that there are fundamental similarities in the structure of NPs and clauses, especially when the NPs are complex derived nominals, e.g. Gödel’s proof of the theorem vs Gödel proved the theorem. The primary correspondence between NPs and clauses in RRG is that both have a layered structure and in both there are operators modifying the layers. In the layered structure of the NP, there is a nominal nucleus and a nominal core consisting of the nucleus and the arguments of a complex derived nominal, e.g. the construction of the bridge by the Mafia-controlled company, but there is only one level, the NP level, corresponding to the clause and sentence levels of the layered structure of the clause. As with the clause, there is a periphery for each level. The operators in the NP are given in Table 1.3, adapted from Rijkhoff (2002). Nominal aspect concerns whether the referent is an individual, parts of an individual, a set of individuals, or a kind; see Jackendoff (1990) for discussion of parallels between the mass/count distinction in NPs and the telic/atelic distinction in verbs. CoreN operators are about quantity, while NP-level operators ground the NP in discourse, analogous to the function of some of the clause-level operators in the clause, especially tense and illocutionary force. A preliminary general schema for the layered structure of the noun phrase is given in Figure 1.17. It follows the layered structure of the clause in having distinct projections for constituents and operators. Syntactic templates for NPs and PPs would be stored in the syntactic inventory, along with the core and other templates discussed in section 1.3. The projection grammar representation of the layered structure of the noun phrase of two English NPs is given in Figure 1.18. The initial the is not attached to anything in the constituent projection, and that reflects its status as one of the NP operators. Operators in the NP follow the same iconic ordering constraint as operators within the clause (Rijkhoff 1990, 2002). Each of the NPs in the complex derived nominal in Figure 1.18 would be given a double projection representation in a complete representation. As in the clause, there are peripheries modifying all three levels. Since the primary function of nuclearN operators is restrictive modification, it is reasonable to suppose that the nuclearN periphery would contain adjunct restrictive modifiers,


NP COREN

(NPIP)

NUC

(NP/ADV)

N

(PP)

(PP)

N NASP

>NUC N > COREN

NUM

> COREN

QNT NEG

>COREN

DEF

> NP

DEIC

>NP

Figure 1.17 The layered structure of the NP (preliminary)

NP CORE N PERIPHERYN >NUCN ADJ the three big

N bridges N NUC N CORE< NUM N

QNT DEF

>CORE N > NP

NP CORE N < NUC N

PERIPHERYN PP

PP

PP

N the construction of the bridge by the company in New York City N NUC N COREN DEF

> NP

Figure 1.18 The layered structure of the NP in English

25

26

syntactic structure

e.g. adjectives and nominal modifiers, and restrictive relative clauses (see section 6.8). Non-restrictive modifiers, on the other hand, would occur in the NP-level periphery (see Figure 6.29). The constituents of the coreN periphery would be the adjunct setting PPs and adverbials of complex event expressions, as in Figure 1.18; this is analogous to their location in the layered structure of the clause. In earlier work in RRG, adjectives were taken to be nuclearN operators, but as pointed out by Jeruen Dery (contribution to RRG discussion list), adjectives would be the only lexical category functioning as an operator in either the clause or the NP. Otherwise, operators are grammatical (functional) rather than lexical categories. Hence adjectives are best treated on the analogy of adverbs in the clause: they are constituents of the (nuclearN ) periphery whose position is constrained by the iconicity principle – they must occur closer to the nominal nucleus than coreN - and NP-level operators and modifiers. It was shown in the previous section that the position of adverbs in a clause is constrained by the operator projection and the iconicity principle regarding operator morphemes and their scope; adjectives are similarly restricted in the NP.7 One of the major parallels between NPs and clauses posited by Chomsky (1970) and Jackendoff (1977) is that both have structurally defined subject positions, and this particular similarity is specifically denied in the RRG account. There are two main reasons for this. First, Comrie and Thompson (1985) show that virtually no other languages allow double genitives analogous to the English ones, e.g. the company’s construction of the bridge, and therefore it is highly inappropriate to base a universal theory of NP structure on a construction which is found in only a single language. In the RRG analysis, the form in Figure 1.18 is posited as the unmarked form for a complex derived nominal. Second, there is considerable evidence within English that the NP in this position is not analogous to a subject in a clause. Nunes (1993) argues that the preposed possessor position is actually functionally similar to the left-detached position in a clause. One piece of evidence for this comes from the fact that non-arguments which cannot be subject in a clause can occur in the preposed possessor position, e.g. Yesterday’s destruction of the city by a hurricane. It also functions similarly to the precore slot, as WH-words can occur in it as well, e.g. which proof of the theorem. This position will be labeled the ‘NP-initial position’ [NPIP] in the NP, and it plays a role in both projections in English at least, as it also marks the definiteness of the NP. In other languages, e.g. Portuguese, Lakhota and Swahili, an article and a possessor phrase may both occur, in which case the NP-initial position is part of the constituent projection only (cf. Haspelmath 1999). A simplified example of an English NP with an NP-initial position is given in Figure 1.19. Possessive pronouns occur in the NP-initial position in English, 7

Another way of expressing this contrast is that operators are closed-class items, not open-class. Matasović (2001) argues that in some languages adjectives are a limited, closed class, and furthermore that attributive adjectives may not head phrases. Hence it may be the case that in some languages attributive adjectives are in fact best treated as a nuclear operator, while in others they should be treated as an ad-nuclearN adjunct modifier, as in Figure 1.18.


NP CORE N < NUC N

NPIP NP

PERIPHERYN PP

PP

N

the company's construction of the bridge in New York City N NUC N COREN DEF

> NP

Figure 1.19 The layered structure of the NP in English with NP-initial position

as do demonstratives in languages in which they are a subtype of pronoun, as in English. In Lakhota the demonstrative occurs as the last element in the NP (default), and therefore there must be a position analogous to the NP-initial position for it to occur in; it will be termed the ‘NP-final position’ [NPFP]. The English NP this book and its Lakhota counterpart, wówapi ki lé [book the this], are given in Figure 1.20; the English demonstrative this occurs in the NP-initial position in (a), while the Lakhota demonstrative lé occurs in the NP-final position in (b). In languages in which demonstratives are non-pronominal and cannot head NPs on their own, they would not be in the NP-initial position or NP-final position; they would be represented solely in the operator projection akin to articles. Possessors receive variable treatment cross-linguistically, ranging from constituents in the NPIP, as in English, to being treated as a special subclass of adjectives, e.g. in Croatian, in which an expression like ‘Ivan’s book’ would not be expressed as a genitive construction (kniga Ivan-a [book Ivan-GEN]) but rather as Ivanova kniga, in which the possessor is a derived adjective. The existence of the NP-final position, as the rightward counterpart of the NP-initial position, provides another parallel between the layered structure of the clause and layered structure of the noun phrase: both have pre- and post-core positions (left-detached position, precore slot, right-detached position, postcore slot in clauses and NP-initial position, NP-final position in NPs). NP

NP

NPIP

COREN

COREN

NPFP

NP

NUC N

NUC N

NP

PRO

DEM

this

PRO

N

N

book

wówapi

N

N

DEF > NP DEIC > NP (a)

DEM

ki

lé

NP < DEF NP

PERIPHERY NPERG NP LOC

NPERG CORE N PRO DEM NUC N

NM.ERG mountains

NUC

COREN

PRED

NUC N NPABS

NP

ABS

COREN NUC N

PRO DEM

N

Baŋgul gambia

CORE

N

V

yaaŋgu bayi

man.ERG NM.ABS

¸durgau

speared

N

N

V

NUC N

NUC N

NUC

COREN

CORE N

CORE

NP

N bargan. wallaby.ABS

N NUC N

CLAUSE< TNS COREN SENTENCE DEIC

DEIC

> NP

> NP

Figure 1.21 The structure of the Dyirbal sentence in (1.22b)

Despite the manifest morphosyntactic differences between the two languages, the comparable structures in each sentence are represented in comparable ways: yaaŋgu and man head NPs functioning as core arguments, as do bargan and wallaby; d¸urgau and speared are verbs functioning as nuclei; and gambia and in the mountains function as locative adjuncts. Moreover, in the NPs headed by yaaŋgu, man, bargan and wallaby there is a determiner which modifies it at the NP-level. The main differences between them lie in two areas: an adjacency condition on NP constituents, which English (along with the vast majority of languages) has but Dyirbal lacks, and grammatical constraints on NP and verb order, which English has and Dyirbal lacks. Hence, the principle in (1.1b) is satisfied, and, as in the comparison of Tzotzil and English in Figure 1.11, this is achieved without making one of the languages look like the other. Thus, the RRG

29

30

syntactic structure SENTENCE CLAUSE CORE
NP CLAUSE< TNS CLAUSE
NP

N

DEF

>NP

IF

SENTENCE

Figure 1.22 The structure of the English translation of (1.22b)

system of syntactic representation allows the analyst to capture important crosslinguistic generalizations about morphosyntactic structure while at the same time permitting the expression of what Sapir (1921) called ‘the structural genius of the language’.8 8

Interestingly, Sapir (1921) seems to have recognized the importance of syntactic templates, which he called ‘sentence-ground-works’: The point that we are really making here is that underlying the finished sentence is a living sentence type, of fixed normal characteristics. These fixed types or actual sentence-ground-works may be freely overlaid by such additional matter as the speaker or writer cares to put on, but they are themselves as rigidly ‘given’ by tradition as are the radical and grammatical elements abstracted from the finished word. New words may be consciously created on the analogy of old ones, but hardly new types of words. In the same way, new sentences are constantly being created but always on strictly traditional lines. (1921:37)

2

Lexical representation and semantic roles

2.0


The next step in the exploration of the syntax, semantics and pragmatics interface is the characterization of the semantic representation of sentences (see Figure 1 in the Introduction). It is based on the semantic representation of the verb or predicating element. Hence much of the discussion in this chapter will be focussed on the lexical representation of verbs, with some limited discussion of the semantic representation of nominals and operators. An equally important facet of the semantic representation of sentences is the semantic relationships which hold between a verb or other predicator and its arguments. The lexicon has become a major component of most contemporary syntactic theories; it is no longer ‘an appendix of the grammar, a list of basic irregularities’ (Bloomfield 1933:274). Grammatical generalizations may be stated both in the syntax and in the lexicon, and therefore the system of lexical representation that a theory uses has a profound effect on the type and nature of the generalizations that may be stated in terms of it.

2.1

Verb classes and logical structures

Underlying any system of lexical representation for verbs and other predicators, implicitly or explicitly, is a theory of verb classes.1 RRG starts from the Vendler (1967) Aktionsart-based classification of verbs into states, achievements, accomplishments and activities, and utilizes a modified version of the representational scheme proposed in Dowty (1979) to capture these distinctions.

1

The classification of verbs and the system of lexical representation presented in this section, despite similar terminology, differ in crucial ways from the ones used in Foley and Van Valin (1984), Van Valin (1990, 1991, 1993b) and other work in RRG prior to the publication of Van Valin and LaPolla (1997). See n. 8. Readers should keep this in mind when looking at those earlier works.

31

32

lexical representation and semantic roles

2.1.1

Verb classes and Aktionsart

Examples of English verbs from each of the Aktionsart classes are given in (2.1). (2.1) a. b. c. d.

States: be sick, be tall, be dead, love, know, believe Achievements: pop, explode, shatter (the intransitive versions) Accomplishments: melt, freeze, dry (the intransitive versions); learn Activities: march, walk, roll (the intransitive versions); swim, think, snow, write, drink

States depict static situations which are inherently temporally unbounded (atelic), and both achievements and accomplishments express changes of state, which are inherently temporally bounded (telic): achievements are instantaneous, while accomplishments are not. Activities are dynamic, inherently temporally unbounded (atelic), states of affairs. Vendler proposed this taxonomy based solely on the analysis of English verbs, and yet it has proved to be of great cross-linguistic validity. Investigations of the following languages have shown that these contrasts are central to the organization of their verbal systems: Lakhota (Foley and Van Valin 1984), Tagalog (Foley and Van Valin 1984), Hausa (Abdoulaye 1992), Sama (Philippines; Walton 1986), Yatye (Kwa, Nigeria; Foley and Van Valin 1984; Stahlke 1970), Tepehua (Totonacan, Mexico; Watters 1988), Italian (Centineo 1996), Georgian (Holisky 1979, 1981a, b), Icelandic (Van Valin 1991), Croatian (Dahm-Draksic 1997), Korean (Yang 1994), Japanese (Hasegawa 1996; Toratani 1998), Mparntwe Arrernte (Australia; Wilkins 1989) and Bribri (Chibchan, Costa Rica; Tomcsányi 1988). These distinctions are also highly relevant to language acquisition (Weist et al. 2004). It would be reasonable to hypothesize that these distinctions are the universal basis of the organization of verbal systems in human language. There is one important non-Vendlerian Aktionsart class, namely semelfactives (Smith 1997). Semelfactives are punctual events which have no result state. Examples are given in (2.2). (2.2) a. b. c. d.

The light flashed. Chris coughed. The tree branch tapped on the window. Dana glimpsed Kim.

There is a derivational relation between two classes which is very important cross-linguistically, namely that between activities and what are called active ac c o m p l is h ments, the telic use of activity verbs. This general pattern relates activity verbs of motion (e.g. run), consumption (e.g. eat) and creation (e.g. paint) to the corresponding active accomplishment verbs. This is illustrated in (2.3) for English.

2.1 Verb classes and logical structures

(2.3) a. a . b. b . c. c .

The soldiers marched in the park. The soldiers marched to the park. Dana ate fish. Dana ate the fish. Leslie painted (for several hours). Leslie painted Mary’s portrait.

Activity Active accomplishment Activity Active accomplishment2 Activity Active accomplishment

There are a few verbs in English that are lexical active accomplishments, e.g. devour and go; that is, they do not alternate with an activity counterpart like the verbs in (2.3). These classes can be characterized in terms of four features, [± static], [± dynamic], [± telic] and [± punctual]. (2.4) a. b. c. d. e. f.

State: Activity: Achievement: Semelfactive: Accomplishment: Active accomplishment:

[+ static], [− dynamic], [− telic], [− punctual] [− static], [+ dynamic], [− telic], [− punctual] [− static], [− dynamic], [+ telic], [+ punctual] [− static], [± dynamic], [− telic], [+ punctual] [− static], [− dynamic], [+ telic], [− punctual] [− static], [+ dynamic], [+ telic], [− punctual]

Most fundamental is the distinction between static and non-static verbs, which distinguishes verbs which code a ‘happening’ from those which code a ‘nonhappening’. In other words, with reference to some state of affairs, one could ask, ‘what happened?’ or ‘what is happening?’ If, for example, a sentence like Bob just ran out the door could be the answer to this question, then the verb run is [− static]. On the other hand, a sentence like John knows Bill well could not be the answer to this question, because nothing is taking place. Hence know is a [+ static] verb. By this criterion activities, achievements, semelfactives, accomplishments and active accomplishments are [− static]. States, however, are [+ static]. The feature ‘dynamic’ refers to whether the situation involves action or not. Activities and active accomplishments both involve action, as indicated by the fact that these verbs can be modified by adverbs like violently, vigorously, actively, strongly and energetically. States, achievements and accomplishments are nondynamic and do not occur with adverbs like these. Some semelfactive verbs do seem to be dynamic, e.g. cough as in He coughed once violently, while others do not, e.g. glimpse as in *He glimpsed the robber strongly. The feature ‘telic’ has to do with whether a verb depicts a state of affairs with an inherent terminal point or not. States and activities lack inherent terminal points. For example, a sentence like Chris is smart makes no reference to a temporal 2

It should be noted that this contrast cannot be reduced to the presence or absence of articles, as claimed in, e.g., Verkuyl (1993), because it occurs in languages which do not have articles, e.g. Georgian, Japanese, Pirahã (Everett 1986) (see Van Valin and LaPolla 1997, section 3.2.1). This contrast revolves around whether the direct object is a specified entity or quantity, in which case it delimits the action and supplies a temporal boundary for it, or whether it is unspecified and therefore does not serve to delimit the action. See Dowty (1979:60ff.) and Jackendoff (1996) for a detailed explication of the semantic basis of these contrasts.

33

34


boundary, and is therefore non-telic (atelic). In The earth is orbiting around the sun, for example, there is a reference to an activity, but orbiting need not terminate. This is not a function of the progressive aspect; in the snow is melting on the driveway, the verb melt entails that there is a terminal point at which the snow will be melted. Therefore, orbit is [− telic], while the intransitive verb melt is [+ telic]. Active accomplishments are telic: in Dana ate the sandwich, there is an inherent terminal point, namely the point at which the sandwich is completely consumed. Similarly, in Pat walked to the park, the inherent endpoint is when Pat arrives at the park. Achievements also have terminal points; if a bomb explodes or a window shatters, the terminal point is the moment of the explosion or the shattering. An achievement is a transition between one state of affairs (the bomb is unexploded, the window is whole) and a new state of affairs (the bomb is exploded, the window is shattered). Hence these verbs are [+ telic] as well. Semelfactives, on the other hand, are pure events with no transition between one state and the next. In, e.g. The light flashed once, the light is in exactly the same state after the event as it was before it; there is no change of state, as with an achievement, hence no transition. Consequently, they are considered to be [− telic]. Therefore, states, activities and semelfactives are unbounded (atelic), while achievements, accomplishments and active accomplishments are bounded (telic). The final feature, [± punctual], distinguishes events with internal duration from those which lack it. The verbs melt and pop can both involve changes of state, as in the ice melted and the balloon popped, but they differ in that the former takes place over a time span, while the latter is instantaneous, for all practical purposes. Hence achievements are punctual, while accomplishments are not. Semelfactives are also punctual, since they denote events without much temporal duration. Since states and activities are atelic, they must by definition involve temporal duration, and therefore they are always [− punctual]. Each of these six classes has a causative counterpart, as exemplified in (2.5). (2.5) a. a . b. b . c. c d. d . e. e . f. f .

State: Causative state: Achievement: Causative achievement: Semelfactive Causative semelfactive Accomplishment: Causative accomplishment: Activity: Causative activity: Active accomplishment: Causative active accomplishment:

The boy is afraid. The dog frightens/scares the boy. The balloon popped. The cat popped the balloon. The pencil tapped on the table. The teacher tapped the pencil on the table. The ice melted. The hot water melted the ice. The soldiers marched in the park. The sergeant marched the soldiers in the park. The soldiers marched to the park. The sergeant marched the soldiers to the park.

There are a number of syntactic and semantic tests for determining the class of a verb. A list of possible tests for the verb class are given in Table 2.1; the causative classes will be discussed below. The ‘*’ means that certain complications arise with this test, which will be discussed below. It should be noted that it is necessary


Table 2.1 Tests for Aktionsart classes Criterion

State

Achiev

Seml

Accomp

Activity

Active Accomp

1. Occurs with progressive 2. Occurs with adverbs like vigorously, actively, etc. 3. Occurs with adverbs like quickly, slowly, etc. 4. Occurs with X for an hour, spend an hour Xing 5. Occurs with X in an hour 6. Can be used as stative modifier

No*

No*

No*

Yes

Yes

Yes

No

No

Some*

No

Yes

Yes

No

No*

No*

Yes

Yes

Yes

Yes*

No*

Yes*

Irrelevant*

Yes

Irrelevant*

No

No*

No*

Yes

No

Yes

Yes

Yes

No

Yes

No

Yes

to adapt the tests to the features of the language under investigation. For example, one of the tests Dowty (1979) gives to differentiate states from the non-states is ‘non-states have a habitual interpretation in simple present tense’, which is clearly an English-specific test. Test 1 is useful only in languages like English, Turkish and Icelandic which have a progressive aspect; it can occur with activities (2.6d), accomplishments (2.6b), active accomplishments (2.6e), but not with most states3 (2.6a, a ) or achievements with a singular subject (2.6c). When it occurs with semelfactives, it necessarily yields an iterative reading (2.6f); this is also the case when the progressive is added to an achievement with a plural subject. 3

There is an additional complexity with the progressive test. Some stative predicates, such as sit, stand and lie, may occur with the progressive under certain circumstances, as in (i). (i) a. b.

The book is lying on the table. The city lies/*is lying at the base of the mountains.

Carlson (1977) calls the statives in (ia) stage-level predicates, because they depict a situation which is not necessarily permanent. When the situation is necessarily permanent, as in (ib), the progressive is impossible. Carlson refers to statives representing permanent states or attributes as individual-level predicates.

35

36

(2.6) a. a . b. c. d. e. f.


*Dana is being tall / fat / a doctor. *Pat is knowing the answer / believing that today is Wednesday. The ice is melting. *The firecracker is popping. (cf. The firecrackers are popping.) Kim is dancing/singing/running/talking/crying/sleeping. Chris is walking to the park. The light is flashing (*once).

This test is marked ‘No*’ for semelfactives, because if an adverb like once or one time is added to make an iterative reading impossible, the progressive is unacceptable. The progressive with a semelfactive verb (or an achievement verb with a plural subject) yields a verb which patterns like an activity verb. Test 2 involves the ability to co-occur with adverbs that code dynamic action, e.g. vigorously, gently, powerfully, etc. This is a test for the feature [dynamic]. (2.7) a. a . b. b . c. c . d. e. e .

*Max is vigorously tall / fat / a linguist. *Max vigorously knows the answer / believes that today is Wednesday. *The snow is melting/melted vigorously. *The window shattered vigorously. Mary is dancing/singing/running/talking/crying vigorously/actively. The house is shaking violently/vigorously/forcefully. Pat ran energetically to the park. Chris coughed once violently. *Chris caught sight of the intruder forcefully/strongly/energetically.

There is an important caution relevant to this test. It is crucial to be careful to avoid adverbs which require a controlling subject, e.g. deliberately, carefully. While they are compatible with many types of activity verbs, they are also incompatible with activity verbs which have subjects which refer to non-agentive participants in the action, e.g. shiver as in the dog shivered violently/*deliberately in the cold, or shake as in the house shook violently/*carefully during the earthquake. Consequently, if such an adverb is acceptable with a verb, then the verb is dynamic, but if it is incompatible with a verb, this may be due to the agentive aspect of these adverbs. Hence it is necessary to also use adverbs which are compatible with involuntary verbs like shiver and with verbs like shake which can have an inanimate subject. If a language lacks adverbs of this kind, then other kinds of adjunct verbal modifiers can be used, e.g. PPs like with great force, in a vigorous way, which express the same thing as these adverbs. Test 3 applies only to non-stative verbs and distinguishes non-punctual from punctual verbs. Adverbs like quickly, rapidly and slowly, which will be called ‘pace’ adverbs, can occur with events involving temporal duration, regardless of whether they involve dynamic action, e.g. the snow is melting slowly/??vigorously, John slowly/*vigorously realized his mistake. The ‘*’ on the ‘No’ in the achievement and semelfactive columns indicates that pace adverbs indicating very short temporal intervals are marginally acceptable with these verbs, e.g. the bomb exploded instantly, the light flashed instantly. Hence with achievement verbs it is necessary to use pace adverbs which indicate a relatively slow process, e.g.


the bomb exploded *slowly/*gradually. Again, with respect to semelfactive verbs these adverbs are possible only on the iterative reading, and therefore adding once should render them unacceptable, e.g. ??The tree branch tapped slowly on the window once. Tests 4 and 5 distinguish telic from non-telic verbs. When applied to other languages, they require one to determine which adposition indicates duration (the for test) and which indicates completion (the in test). Test 4 isolates the property of having duration in time; it shows that states, accomplishments, activities and active accomplishments all have temporal duration, but achievements do not. Semelfactives can take a for PP expressing very short duration, as in (2.8f). Test 5 focuses on terminal points. If something is done in ten minutes, then explicit reference is being made to the termination point of the event. In other words, the event started at a certain time and ended ten minutes later. But if something is done for ten minutes, the same event could still be going on at a later time. All the for-phrase indicates is that an event went on for a certain amount of time, without any information about when it began or when it ended. So in he read the book in an hour, the event began and ended in the space of one hour, with the subject having finished reading the book, whereas in he read the book for an hour, there is no indication of when the action began or ended, and the same event could still be going on at a later time. In general, states and activities readily take for-phrases, while accomplishments take in-phrases. Because achievements are punctual, they are only compatible with in-phrases referring to an exceedingly short period of time, e.g. in the blink of an eye, in an instant, in a fraction of a second. They are incompatible with in-phrases referring to temporal periods longer than this, e.g. in ten seconds, in a minute and in an hour, unless they have an iterative reading, and accordingly they are marked ‘No*’ in Table 2.1. Hence this test should also be used with temporal expressions of substantial duration. (2.8) a. a . b. c. d. e. e . f. f .

Max was tired/ill/happy for/*in an hour. Max liked Susan for/*in an hour. The snow melted in/for an hour. The window shattered in/*for a fraction of a second. (*The window shattered in an hour.) Mary danced/sang/cried/talked/slept for/*in ten minutes. Tom drank the glass of beer in/*for an hour. Tom drank beer for/*in an hour. The light flashed once for/??in an instant. *The light flashed once in/for an hour.

State predicates which code inherent properties do not normally take for phrases, e.g. *Sandy was tall/thin/short/fat for an hour. Hence there is an asterisk on the ‘Yes’ indicating that this test is problematic for some state predicates (cf. n. 3). Some accomplishments and active accomplishments can take forphrases, e.g. the clothes dried for ten minutes or Chris ran to the park for five minutes, which follows from their being non-punctual, which is the main point

37

38


of test 3. Hence the occurrence of for-phrases with accomplishments is really redundant and tells us nothing new about accomplishments or active accomplishments. Hence it is marked as ‘irrelevant’ in Table 2.1 for them. Finally, there is an additional co-occurrence which must be noted. Achievements, semelfactives and activities do co-occur with in-phrases, e.g. The bomb will explode in one hour, Mary will sing in ten minutes; these phrases refer to the time until the onset of the action or event, not to the temporal duration of the event itself and are therefore irrelevant to these tests. Thus, it is not sufficient simply to ascertain the type of temporal phrase that a verb can occur with; it is, rather, necessary to pay attention to the meaning of the sentence as well. Test 6 serves primarily to distinguish the two punctual types from each other. As noted earlier, semelfactives have no result state, and therefore they cannot be used as stative modifiers, e.g. *the tapped window, *the flashed light. Achievements, on the other hand, do have a result state and therefore can be used as stative modifiers, e.g. the shattered window, the burst blood vessel. This is related to the fact noted earlier that semelfactives can have an iterative interpretation with a singular subject, while achievements can only have such a reading with a plural subject. Because the subject of an achievement undergoes a change of state, it cannot undergo it again, and therefore a different referent is required for the action to repeat, whereas, because the subject of a semelfactive verb does not undergo a change of state, it can repeat the action, hence the possibility of an iterative reading with a singular subject. This contrast could be used as an additional test for differentiating achievements from semelfactives. There is no simple syntactic test to determine whether a verb is inherently causative or not, but paraphrases can be useful, as illustrated in (2.9). (2.9) a. b. c. d.

The dog caused the boy to be afraid. The cat caused the balloon to pop. The hot water caused the ice to melt. The girl caused the ball to bounce around the room.

It is important to make sure that the paraphrases have the same number of NPs as the original sentence being paraphrased; that is, ‘Pat causes Chris to come to have the book’ is an appropriate paraphrase of Pat gives the book to Chris, but ‘Leslie causes Leslie/herself to run’ is not a possible paraphrase of Leslie runs (it is at best a paraphrase of Leslie made herself run). This means that this test cannot apply to single argument verbs, i.e. verbs that have one argument in their basic form, because it would be impossible to make a causative paraphrase with a single participant. It must be emphasized that the claim here is not that The dog scared the boy and The dog caused the boy to be afraid mean exactly the same thing; rather it is that if The dog scared the boy is true, then The dog caused the boy to be afraid is also true; the converse need not hold.4 Furthermore, the causative relation intended here is exactly the same one that is signalled by causative morphology 4

In discussions of verb semantics and paraphrases, it has often been pointed out that there are situations under which John caused Bill to die would be true but John killed Bill would not be.


39

in many languages, e.g. Lakhota in´ıha¸ ‘be scared, frightened, amazed, awed’ vs in´ıha¸-ya (-ya ‘cause’) ‘scare, frighten, amaze, awe’; Barai (Papua-New Guinea; Olson 1981) mae ‘be happy’, visi ‘be sick’, vs ma-d- [happy+cause] ‘please’ and visi-nam- [sick+cause] ‘sicken’. Table 2.2 Tests for determining predicate classes Class State Achievement Semelfactive Accomplishment Activity Active accomplishment Causative state Causative achievement Causative semelfactive Causative accomplishment Causative activity Causative active accomplishment

Test 1 Prog.

Test 2 Dynamic

Test 3 Duration

Test 4 for-PP

Test 5 in-PP

Test 6 Stat.Mod.

Test 7 Cause

No* No* No* Yes Yes Yes

No No No* No Yes Yes

No No* No* Yes Yes Yes

Yes* No* Yes* Irrelev.* Yes Irrelev.*

No No* No* Yes No Yes

Yes Yes No Yes No Yes

No No No No No No

Yes* No

Yes* Yes*

No No*

Yes No

No No*

Yes Yes

Yes Yes

No*

Yes*

No*

No*

No*

No

Yes

Yes

Yes*

Yes

Irrelev.*

Yes

Yes

Yes

Yes Yes

Yes Yes

Yes Yes

Yes Irrelev.*

No Yes

Yes Yes

Yes Yes

When the causative versions of the classes listed in Table 2.1 are included, the result is given in Table 2.2. The causative paraphrase test is test 7. The ‘*’ for achievement and causative achievements and semelfactives with respect to tests 3 and 5 is the same as discussed above, as is the ‘*’ for test 4 with state predicates. Causative states present some interesting complexities with respect to tests 1 and 2. Specifically, the more active the causing state of affairs is, the better the progressive and dynamic adverbs are with causative state predicates. Consider the following contrasts. (2.10) a. a . b. b . c. c .

Your attitude upsets / ?is upsetting me. Your boorish behaviour upsets / is upsetting me. Your clothes nauseate / ?are nauseating me. The smell of your clothes nauseates / is nauseating me. The clown’s funny hair amuses / ?is amusing the children. The clown’s zany antics amuse / are amusing the children.

This is correct but irrelevant. The crucial point is that in all circumstances in which John killed Bill would be true, John caused Bill to die would also be true. This is the paraphrase relation that is essential for this test.

40


The first sentence in each pair presents a rather static situation as the cause of the state of affairs, while the second presents a more dynamic causing state of affairs. While none of the combinations is impossible, the progressive is better with the more dynamic causing state of affairs and worse with the more static one. Dynamic adverbs also force a dynamic reading for the causing state of affairs. For example, the sentence the clown actively amused the children could only be a report about the state of affairs described by (2.10c ), not (2.10c). The ‘Yes*’ for test 2 for causative achievements, semelfactives and accomplishments reflects the fact that this type of adverb is not always acceptable with these verbs. It modifies the causing activity. Because they are sometimes acceptable, causative accomplishments differ little from causative active accomplishments in terms of these tests. But there are important differences. First, there should always be at least some dynamic adverbs compatible with causative active accomplishments, and because there are two activities, the causing activity and the caused activity, there may be ambiguity as to which one is being modified, something which is not the case with causative accomplishments, in which a dynamic adverb could be interpreted only as modifying the causing activity. Second, causative accomplishments are ultimately related to a state predicate, whereas causative active accomplishments are ultimately related to an activity predicate. That is, the non-causative form of a causative accomplishment should be an accomplishment, which should involve a specific result state. The non-causative form of a causative active accomplishment, on the other hand, should be an active accomplishment, which should involve a specific activity. It should also be noted that causative accomplishments are much more common than causative active accomplishments, and therefore in unclear cases it is more likely that the verb would be a causative accomplishment rather than a causative active accomplishment. These tests are not perfect, but taken together they enable the analyst to distinguish the classes. As noted above, it is necessary to adapt the tests to the language being investigated, and not all of them are equally useful. If a language lacks a progressive aspect, for example, then test 1 is irrelevant. Finally, it is necessary to be sensitive to what may be called ‘local co-occurrence effects’ in interpreting the tests. For example, suppose one applied test 3 to the English verb rush, as in She rushed across the room, in order to determine whether this verb has temporal duration or not, yielding She rushed quickly/swiftly/*slowly across the room. Some but not all pace adverbs are possible here; what is one to conclude? The correct conclusion is that rush has temporal duration and therefore is either an active accomplishment or activity verb. But what about the incompatibility with slowly? This is an example of a local co-occurrence effect; because part of the inherent meaning of rush is to do something with some degree of rapidity, slowly conflicts with this aspect of the meaning of rush. This is not due to the verb not having temporal duration, as its co-occurrence with quickly and swiftly show. Rather, the incompatibility of rush and slowly is due to an aspect of the meaning of rush which is unrelated to what test 3 is testing for. In the same vein, it is possible that only one of the class of adverbs of the type mentioned in test 2


41

is compatible with a particular verb; that would be sufficient to show that the verb rates a ‘yes’ for the test. Other factors irrelevant to the point of the test may cause the other adverbs to be ruled out. Thus one must be sensitive to these local co-occurrence effects in interpreting the results of the tests. In many languages, verbs in these different classes may be overtly morphologically related to each other. Consider the following examples from Huallaga Quechua (Weber 1989). (2.11) a.

State qarwashyellow ‘be yellow’

Accomplishment qarwash-ta:yellow-become ‘become yellow’

Causative accomplishment qarwash-ta:-chiyellow-become-cause ‘make something yellow’

b.

hanqahanqa-ya:above.on.slope above.on.slope-become ‘above with respect to slope’ ‘become higher’

hanqa-ya:-chiabove.on.slope-become-cause ‘make something higher’

c.

hatunbig ‘be big’

hatun-ya:big-become ‘become bigger’

hatun-ya:-chibig-become-cause ‘make something bigger’

d.

umasapabig.headed ‘be big-headed’

umasapa-ya:big.headed-become ‘become big-headed’

umasapa-ya:-chibig.headed-become-cause ‘make someone big-headed’

e.

—

wan¯udie ‘die’

wan¯u-chidie-cause ‘kill’

f.

—

yachalearn ‘learn’

yacha-chilearn-cause ‘teach’

In (2.11a–d) accomplishment verbs are formed from state predicates by the addition of the suffix-ya:- ‘become’, and causative accomplishments are formed from them by the addition of the causative suffix -chi-. As (2.11e–f) show, -chi- can be added to underived accomplishment verbs as well. A rather different pattern expressing the same relationships obtains in Yagua (Peru; Payne and Payne 1989), Russian and French. (2.12) a. b. c.

Causative Accomplishment Yagua -muta- ‘open [tr]’ French briser ‘break [tr]’ Russian razbit’ ‘break [tr]’

Accomplishment -muta-y- ‘open [intr]’ se briser ‘break [intr]’ razbit’sja ‘break [intr]’

State -muta-y-maa ‘be open’ brisé ‘broken’ razbitij ‘broken’

In these three languages, the base form of the verb is a causative accomplishment, and the accomplishment and state forms are derived morphologically from it. This is a very common pattern cross-linguistically; see Haspelmath (1993) for a survey of the morphological patterns exhibited by verbs in this alternation. It is also the case that some languages overtly mark the contrast between activity verbs and active accomplishment verbs, e.g. Russian est’ ‘eat [activity]’ vs s”est’ ‘eat [active accomplishment]’, Georgian c’er ‘write [activity]’ vs dac’er ‘write

42


[active accomplishment]’. In all of these languages there are overt morphological relationships among some of the different classes.

2.1.2

Logical structures

An explanation for these patterns can be found in the lexical representations used in RRG: verbs are analysed in terms of a lexical decomposition system in which state and activity predicates are taken as basic and the other classes are derived from them. States are represented as bare predicates, e.g. know (x, y), dead (x). Activity verb representations all contain the element do , e.g. do (x, [cry (x)]) ‘cry’, do (x, [eat (x, (y))]) ‘eat’. Achievements, which are punctual changes of state or onsets of activity, are represented as a state or activity predicate plus an INGRessive operator, e.g. INGR shattered (x) ‘shatter [intr]’. English does not have any lexical verbs indicating a punctual onset of an activity, but other languages do. In Russian, plakat’ ‘cry’ is an activity verb, and zaplakat’ ‘burst out crying’ is an achievement, i.e. INGR do (x, [cry (x)]); note the translation using the punctual expression burst out. Semelfactives likewise can be based on states or activities, e.g. glimpse would have the representation SEML see (x, y), while cough would have the representation SEML do (x, [cough (x)]). This captures the fact that only semelfactives based on activities have an activity reading when iterative, e.g. Pat is coughing vs *Chris is glimpsing Kim. This also accounts for their differential behaviour with dynamic adverbs: those based on activities can take such adverbs, e.g. He coughed once violently, while those based on states cannot, *He glimpsed her vigorously. Accomplishments, which are non-punctual changes of state or onsets of activity, are represented as a state or activity predicate plus a BECOME operator, e.g. BECOME melted (x) ‘melt [intr]’. Russian also provides an example of a non-punctual onset of activity verb: govorit’ ‘talk, speak’ is an activity verb, and zagovorit’ ‘start talking’ is an accomplishment, i.e. BECOME do (x, [speak (x)]). Causative verbs have a complex structure consisting of a predicate indicating the causing action or event, usually an activity predicate, linked to a predicate indicating the resulting state of affairs by an operator-connective CAUSE, e.g. [do . . .] CAUSE [BECOME pred . . .].5 Accomplishment verbs like melt involve both a process that takes place over time, and an inherent endpoint of the process leading to a result state. Achievement 5

Treating causatives as all having the same ‘CAUSE’ element is a gross oversimplification for many reasons, only two of which will be mentioned here. First, there is a contrast among three basic types of causality: (1) Pam made Sally go [Direct (Coercive)]; (ii) Pam had Sally go [Indirect (Non-coercive)]; and (iii) Pam let Sally go [Permissive]. Both direct and indirect causality will be represented by ‘CAUSE’, and permissive causality will be represented by ‘LET’ in logical structure. English verbs like let, drop and release would have LET instead of CAUSE in their logical structure. Virtually all the examples to be discussed involve direct or indirect causality. Second, there is an important contrast between implicative and non-implicative causality; that is, the difference revolves around whether the resulting state of affairs is necessarily entailed or not. In English, lexical causatives and direct causality are implicative, whereas permissive causality is not: *Pam broke the dish, but it didn’t break; *Pam made Sally go, but she did not go; Pam let Sally go,


verbs like pop, on the other hand, have no process, only a punctual event leading to a result state. Thus, an accomplishment can be analysed as a process plus an achievement, if the final moment of the process is equated with the punctual event of the achievement. There is no direct representation of such a process in this decomposition system. It is not the same as an activity, since it is not dynamic, i.e. such processes do not co-occur with the adverbs of test 2 in Table 2.1. This can be seen in English when verbs like melt occur in the progressive, which isolates the process from the endpoint, as in (2.13). (2.13) a. b.

The ice is melting. The ice is melting slowly/*vigorously.

There are languages in which verbs directly express processes with no necessary implication of an endpoint and result state, unlike English verbs like melt and dry. The following examples are from Mparntwe Arrernte (Wilkins 1989). (2.14) a.

Ayenge irrernte ne-ke. 1sgNOM cold COP-PAST ‘I was cold.’

b.

Ayenge irrernt-irre-ke. 1sgNOM cold-PROC-PAST ‘I got colder/cooler/*cold.’

b .

Ayenge iparrpele/*tyepetyepele irrernt-irre-ke. 1sgNOM quickly/energetically cold-PROC-PAST ‘I got colder quickly/*energetically.’

c.

Ayenge irrernte-arle-irre-ke. 1sgNOM cold-RES-PROC-PAST ‘I got cold.’

but she did not go. In some languages, e.g. Japanese (Ikegami 1985) and Korean (Park 1993), lexical and direct causatives are not implicative. The following examples are from Park (1993). (1) a.

b.

Chelswu-nun mwul-ul el-li-ess-una, mwul-i an-el-ess-ta. -TOP water-ACC freeze-CAUS-PAST-but water-NOM NEG-freeze-PAST-DEC ‘Chelswu froze the water, but the water did not freeze.’ Congi-lul thay-wu-ess-una, ku congi-ka an-tha-ess-ta. paper-ACC burn-CAUS-PAST-but the paper-NOM NEG-burn-PAST-DEC ‘I burned the paper, but the paper didn’t burn.’

In order to make them implicative, the verb noh ‘put’ must be serialized with them, and this creates a contradiction within the sentence frame of (2). (2) a.

b.

*Chelswu-nun mwul-ul el-li-e-noh-ess-una, -TOP water-ACC freeze-CAUS-LNK-put-PAST-but an-el-ess-ta. NEG-freeze-PAST-DEC ‘Chelswu froze the water, but the water did not freeze.’ *Congi-lul thay-wu-e-noh-ess-una, ku congi-ka paper-ACC burn-CAUS-LNK-put-PAST-but the paper-NOM ‘I burned the paper, but the paper didn’t burn.’

mwul-i water-NOM

an-tha-ess-ta. NEG-burn-PAST-DEC

For discussion of these issues, see, e.g., Jackendoff (1990), Talmy (2000), especially with respect to the latter’s theory of force dynamics.

43

44


The crucial contrast is between (2.14b) and (c): the suffix -irre added to the stative stem irrernte ‘cold’ means ‘become colder’ or ‘become cooler’ but not ‘become cold’, i.e. it signals a change from less cold to more cold without entailing that the process has reached the endpoint of being cold. Like (2.13b) in English, a dynamic adverb like tyepetyepele ‘energetically’ is impossible with a process. In order to signal the reaching of the endpoint, the suffix -arle ‘result’ must be added, as in (c), to indicate that the process reached its termination yielding a result state. Hence in Mparntwe Arrernte it is necessary to represent processes independent of a possible endpoint and result state. In order to do this, it is necessary to introduce an operator PROC for ‘process’. The decompositional representation for the examples in (2.14) would be as in (2.15). (2.15) a. b. c.

cold (1sg) PROC cold (1sg) PROC cold (1sg) & INGR cold (1sg)6

irrernte irrernte + irre irrernte + arle + irre

Thus, BECOME with a state predicate is decomposable into PROC & INGR; ‘&’ is a connective meaning ‘and then’ (see Dowty (1979:74–5)). However, it will continue to be used with verbs like English melt and dry, because they normally entail both a process and a result state.7 Unlike INGR, SEML and BECOME, PROC does not occur with activity verbs. The other three entail an event or transition which is the onset of the activity, and because PROC has no event or transition entailment, it could not be used to characterize some kind of pre-onset process. As shown in (2.3), active accomplishments are composed of an activity predicate plus a change of state which renders them telic. There are two kinds of active accomplishments: those involving verbs of motion, as in (2.3a ), and those involving verbs of consumption and creation as in (2.3b ,c ). With the first kind, the change is a change of location, namely the motion is completed when the subject arrives at a particular location. Hence the decompositional representation of (2.3a ) would be as in (2.16a). With respect to verbs of consumption and creation, on the other hand, the result state is either the consumption or creation of an object, which involves a change of state rather than location. The representations for the examples in (2.3b ,c ) are given in (2.16b,c). (2.16) a. b. c.

do (soldiers, [march (soldiers)]) & INGR be-at (park, soldiers) do (Dana, [eat (Dana, fish)]) & INGR consumed (fish) do (Leslie, [paint (Leslie, Mary’s portrait)]) & INGR exist (Mary’s portrait)

Because active accomplishments are composed of an activity + termination with result state, they are more accurately characterized as ‘active achievements’. 6

7

The INGR operator here is not directly signalled by the verbal morphology, which indicates a process plus a result state. But the existence of a result state necessarily entails a change of state leading to the result, hence the INGR. The effect of the progressive on verbs like this can now be easily expressed: progressive + BECOME pred (x) yields PROC pred . In other words, the progressive cancels the INGR component of BECOME, leaving only PROC. This follows from the incompatibility of the progressive and achievements, as expressed in test 1.


Table 2.3 Lexical representations for Aktionsart classes8 Aktionsart class

Logical structure

STATE ACTIVITY ACHIEVEMENT

predicate (x) or (x, y) do (x, [predicate (x) or (x, y)]) INGR predicate (x) or (x, y), or INGR do (x, [predicate (x) or (x, y)]) SEML predicate (x) or (x, y) SEML do (x, [predicate (x) or (x, y)]) BECOME predicate (x) or (x, y), or BECOME do (x, [predicate (x) or (x, y)])

SEMELFACTIVE ACCOMPLISHMENT

ACTIVE ACCOMPLISHMENT do (x, [predicate1 (x, (y))]) & INGR predicate2 (z, x) or (y) CAUSATIVE α CAUSE β, where α, β are logical structures of any type

However, they will still be referred to as ‘[active] accomplishments’, since this is the standard term for them in the literature. These decompositional representations of verbs are termed ‘logical structures’, and the schemas for the classes are given in Table 2.3. Following the conventions of formal semantics, constants (which are normally predicates) are presented in boldface followed by a prime, whereas variable elements are presented in normal typeface. The elements in boldface + prime are part of the vocabulary of the semantic metalanguage used in the decomposition; they are not words from any particular human language, despite their obvious resemblance to English words. Hence the same representations are used for all languages (where appropriate), e.g. the logical structure for Lakhota t’á and English die would be BECOME dead (x). The variables are filled by lexical items from the language being analysed; for example, the English sentence ‘the dog died’ would have the logical structure BECOME dead (dog), while the corresponding Lakhota

8

The following table compares the verb class taxonomy presented here with that assumed in work prior to Van Valin and LaPolla (1997) (VV&LP). Semelfactive verbs were not included in the 1997 classification. Verb class (1997)

Pre-VV&LP label

State Activity Achievement Accomplishment Active accomplishment Causative state Causative activity Causative achievement Causative accomplishment Causative active accomplishment

State Activity Achievement (punctual) Achievement (durative) Accomplishment — — Accomplishment Accomplishment —

45

46


sentence sˇu´¸ka ki t’é ‘the dog died’ would have the logical structure BECOME dead (ˇsu´¸ka). Thus, the constants should be from the semantic metalanguage, while the variable(s) should be filled by lexical items from the language in question.9 The derivational relationships illustrated in (2.11) and (2.12) can be readily accounted for in terms of this system of lexical decomposition. The state → accomplishment → causative accomplishment pattern found in Huallaga Quechua follows directly from the lexical representations, e.g. in (2.11c), hatun(big (x)) → hatun-ya:- (BECOME big (x)) → hatun-ya:-chi- (. . . CAUSE [BECOME big (x)]). The pattern in Yagua, French and Russian also indicates a systematic relationship among these classes, but the function of the morphological markers is to cancel part of the logical structure rather than to add components to it, e.g. in (2.12a), -muta- (. . . CAUSE [BECOME open (x)]), -muta-y- (BECOME open (x)), in which -y- cancels the ‘. . . CAUSE’ part of the logical structure, and -muta-y-maa (open (x)), in which -maa cancels the BECOME part of the logical structure. This system of lexical representation, while much richer than that employed by some other theories,10 is nevertheless only a first approximation to the kind of decompositional system which is required for deeper lexical semantic analysis. Van Valin and Wilkins (1993) explores the features of such a system, focussing on the decomposition of cognition verbs in English and Mparntwe Arrernte, and Van Valin and LaPolla (1997) explores the extended decomposition of English verbs of saying. Mairal and Faber (2002) investigates the extended decomposition of verbs of cutting in English, while González Orta (2002) examines the semantic structure of the Old English verb secgan ‘say’. Examples of some English verbs with their logical structure are presented in (2.17). (2.17) a.

b.

9

States Pat is a fool. The cup is shattered. Kim is in the library. Dana saw the picture.

be (Pat, [fool ]) shattered (cup) be-in (library, Kim)11 see (Dana, picture)

Activities The children cried. Carl ate pizza.

do (children, [cry (children)]) do (Carl, [eat (Carl, pizza)])

If the constants were also to be drawn from different languages, then the logical structures would become unintelligible and difficult, if not impossible, to compare, e.g. English BECOME dead ´ t’a (ˇsu¸´ka). The point of the semantic (dog) vs German WERDEN tot (Hund) vs Lakhota UYA metalanguage is to have a common representation for the meaning of verbs and other predicators across languages, in order to facilitate cross-linguistic comparison and generalization. 10 The system of lexical decomposition proposed in Rappaport Hovav and Levin (1998) is quite similar to the RRG system, especially to the earlier system referred to in fn. 8, with reference to which the differences are largely notational. Their lexical templates correspond to the logical structure types in Table 2.3. 11 The order of arguments in the logical structure will be discussed in section 2.4.1 below.


c.

d.

e.

Achievements The window shattered. The balloon popped.

INGR shattered (window) INGR popped (balloon)

Se me l factives Dana glimpsed the picture. Mary coughed.

SEML see (Dana, picture) SEML do (Mary, [cough (Mary)])

Accomplishments The snow melted. Mary learned French.

BECOME melted (snow) BECOME know (Mary, French)

47

f.

Active Accomplishments Chris ran to the park. do (Chris, [run (Chris)]) & INGR be-at (park, Chris) Carl ate the pizza. do (Carl, [eat (Carl, pizza)]) & INGR consumed (pizza)

g.

Causatives The dog scared the boy. Max melted the ice. The cat popped the balloon. Sam flashed the light. Felix bounced the ball. Mary fed the pizza to the child.

[do (dog, Ø)] CAUSE [feel (boy, [afraid ])] [do (Max, Ø)] CAUSE [BECOME melted (ice)] [do (cat, Ø)] CAUSE [INGR popped (balloon)] [do (Sam, Ø)] CAUSE [SEML do (light, [flash (light)])] [do (Felix, Ø)] CAUSE [do (ball, [bounce (ball)])] [do (Mary, Ø)] CAUSE [do (child, [eat (child, pizza)]) & INGR consumed (pizza)]

In (2.17g), ‘do (x, Ø)’ represents an unspecified activity. A crucial point to be emphasized is that it is necessary to distinguish the basic lexical meaning of a verb, e.g. drink as an activity verb, from its meaning in a particular context, e.g. drink a glass of beer as an active accomplishment predication. The former would have ‘do (x, [drink (x, y)])’ as its representation in its lexical entry in the lexicon, whereas the latter would have ‘do (x, [drink (x, y)]) & INGR consumed (y)’ as the semantic representation of the core of the clause in which drink appears. A given logical structure is intended to represent a particular meaning or interpretation of a lexical item; it is not necessarily the case that there is a single logical structure underlying all of the uses of a particular verbal lexical item. It is not necessary to list each of these verbs separately in the lexicon; rather, the activity forms would be listed and the active accomplishment use would be derived by the following lexical rules. (2.18) a. b.

Motion verbs: do (x, [pred (x)]) do (x, [pred (x)]) & INGR be-LOC (y, x) Creation/consumption verbs: do (x, [pred1 (x, y)]) do (x, [pred1 (x, y)]) & INGR pred2 (y)

Similarly, verbs which are related morphologically could also be derived by lexical rules. For example, the Huallaga Quechua accomplishment forms in the second column in (2.11) would be derived from the state forms in the first column by an inchoativization rule in the lexicon, and the causative accomplishment forms in the third column would be derived from the accomplishment forms in the second

48


column by a causativization rule in the lexicon.12 For a polysemous verb each meaning would be associated with a different logical structure; for example, take in the sense of ‘obtain’ or ‘get’ would have a different logical structure from take in the sense of ‘carry’. Logical structures are associated fundamentally with the meanings which verbs express. Adjectives are treated as state predicates in logical structure when they function predicatively, and following Schwartz (1993) attributive and identificational predications are given the logical structures in (2.19a ,b ). Specificational predications have the logical structure in (2.19c ), following Pavey (2004). Equational predications are represented as in (2.19d ). (2.19) a. a . b. b . c. c . d. d .

Pat is tall. be (Pat, [tall ]) Kim is a lawyer. be (Kim, [a lawyer ]) Chris is the winner. be (Chris, [the winner]) Kim’s sister is Sandy’s lawyer. equate (Kim’s sister, Sandy’s lawyer)

Attributive Identificational Specificational Equational

Be in these logical structures does not correspond to English be, which, as noted in section 1.3, is not part of the predicate in a copular construction; rather be serves only to indicate attributive, identificational and specificational predications in logical structures. Result state predications do not contain it, e.g. dead (dog) for The dog is dead. In attributive and identificational predications, the second argument position of be is filled by an adjectival or nominal predicate. In a specificational predication, it is filled by a referring expression, which serves as the nucleus of the clause; in such predications, the first argument represents a variable and the second its value (DeClerck 1988). In equational predications, on the other hand, the two argument positions are both filled by referring expressions, and referential identity is asserted. The English verb be is the actual predicate in the nucleus only in equational predications; in the other three types it is an auxiliary verb in the nucleus but not part of the predicate, as in Figure 1.6. English uses be for a wide range of stative predications, while other languages make important semantic distinctions within this range. In Tagalog, for example, inherent attributes and result states are distinguished, as illustrated in (2.20) (from Ricardo Nolasco, personal communication). (2.20) a.

b.

12

Puti ang damit niya. white NOM clothes her ‘Her clothes are white.’ Ma-puti ang damit niya. -white NOM clothes her ‘Her clothes are clean.’

See Van Valin and LaPolla (1997), section 4.6, Van Valin (2004) and section 5.2 below for a discussion of lexical rules in RRG.


The logical structure for (2.20a) is be (damit niya, [white ]), an individual-level predicate (see n. 3), while the logical structure for (b) is white (damit niya), a stage-level predicate. Spanish uses four different verbs in constructions whose English equivalents all use be. (2.21) a. a . b. b . c. c . d. d .

Pat is tall. Pat es alta. Pat is in the library. Pat está en la biblioteca. Pat is cold (i.e. feels cold) Pat tiene frio. The weather is cold. Hace frio.

be (Pat, [tall ]) Spanish: ser ‘to be’ be-in (Pat, library) Spanish: estar ‘to be located’ feel (Pat, [cold ]) Spanish: tener ‘to have’ cold (weather) Spanish: hacer ‘to make, do’

Thus, simply because a number of stative predications in a language all contain the same copula, it cannot be assumed that they have the same logical structure. Predicative prepositions are also state predicates, as exemplified in (2.17a). When predicative PPs function as adjunct modifiers, they take the logical structure of the main verb as one of their arguments. This is exemplified in (2.22). (2.22) a. b.

Chris ran in the park. be-in (park, [do (Chris, [run (Chris)])])

In (2.22a), Chris’ running takes place in the park, and therefore the logical structure of the predicative preposition in is the highest predicate in the logical structure; it takes the park and the logical structure for run as its two arguments. This contrasts with the logical structure of the active accomplishment Chris ran to the park in (2.17f); there the PP expresses the location of the referent of Chris, not the location of the event of running, and in this example to the park is an argument-adjunct PP. Adverbs are represented in logical structure as one-place predicates, which modify different parts of a logical structure.13 Temporal adverbs take the whole logical structure as their argument, as in (2.23). (2.23) a. b.

Yesterday Chris ran to the park. yesterday (do (Chris, [run (Chris)]) & INGR be-at (park, Chris))

Manner adverbs modify activity predicates primarily, while pace adverbs can modify any kind of durational predicate; they are illustrated in (2.24a). Aspectual adverbs modify the basic state or activity predicate, as in (2.24b). (2.24) a. a . b. b .

13

Pat elegantly closed the door slowly. [elegant (do (x, Ø))] CAUSE [slow (BECOME closed (y))] The ice melted completely / The ice completely melted. BECOME (complete (melted (ice)))

See Van Valin and LaPolla (1997), section 4.4.1.2, for a detailed discussion of adverbs and their semantic representation.

49

50

2.2


Operators

Operators like tense, aspect, modality and illocutionary force are very complex semantically, and no attempt will be made to develop a substantive semantic representation for them here. Rather, a place for them in the semantic representations will be provided, so that their interaction with other elements of the representations can be represented. The RRG semantic representation would ultimately have to be given a full interpretation in a formal semantic theory. In order to distinguish operators from the other elements in semantic representations, they will be represented in italicized caps inside angled brackets indicating their scope in logical structure. The general schema is summarized in (2.25), with values arbitrarily chosen solely for illustration purposes. There is a range of values for each operator, which depends on the operator system in the language in question; for example, in a language with a past/non-past tense system, there are two values for the tense operator, whereas in a language with a past/present/future system, there are three values. (2.25)

IF DEC EVID HS TNS PAST STA IRR NEG Ø MOD OBLG EVQ SG DIR Ø ASP PERF LS

A partial semantic representation for Has Kim been crying? would be as in (2.26). (Operators with no specification will be omitted for ease of presentation.) (2.26)

2.3

IF INT TNS PRES ASP PERF PROG do (Kim, [cry (Kim)])

Nouns and noun phrases

Nouns have a semantic representation based on the qualia analysis developed in Pustejovsky (1995) in his Generative Lexicon theory. A non-derived noun like cat or table does not have a logical structure like a verb or predicative preposition, but it does have semantic properties which contribute in an important way to the meaning of a sentence. Consider the following example. (2.27)

Dana began a new novel.

This sentence can normally be construed to mean that Dana began either to read a novel or to write a novel. Where does this meaning come from? Why can’t one interpret this to mean that Dana began to eat a novel, the way one can construe Chris started in on a souvlaki? The answer obviously is that the interpretation derives from the different object NPs in the two sentences, novel, on the one hand, and souvlaki, on the other. In order to capture these facts, Pustejovsky (1991, 1995) proposes the qualia theory of the semantics of nominals. It is summarized in (2.28).

2.3 Nouns and noun phrases

(2.28) a.

Qualia Theory (Pustejovsky 1991:426–7) Constitutive Role: the relation between an object and its constituents, or proper parts 1. Material 2. Weight 3. Parts and component elements

b.

Formal Role: that which distinguishes the object within a larger domain 1. Orientation 2. Magnitude 3. Shape 4. Dimensionality 5. Color 6. Position

c.

Telic Role: purpose and function of the object 1. Purpose that an agent has in performing an act 2. Built-in function or aim that specifies certain activities

d.

Agentive Role: factors involved in the origin or “bringing about” of an object 1. Creator 2. Artifact 3. Natural kind 4. Causal chain

Pustejovsky gives the following representation for novel.14 (2.29) a. b. c. d.

novel (y) Const: narrative (y) Form: book (y), disk (y) Telic: do (x, [read (x, y)]) Agentive: artifact (y), do (x, [write (x, y)]) & INGR exist (y)

The source of the two interpretations for (2.27) is now clear: one reading is based on the telic role of novel, while the other is derived from the agentive role. In the case of (2.27), Dana began a new novel, we would have the following logical structure.15 (2.30)

BECOME do ([Dana (x), {. . .}], [verb ([Dana (x), {. . .}], [novel (y), {. . . , QT [do (x, [read (x, y)])], QA [do (x, [write (x, y)])]}])])

Begin is treated semantically like a complement-taking predicate with an unspecified complement verb (the ‘verb ’ in the logical structure), and the semantic content of the unspecified verb is supplied by the logical structure in the telic 14 15

The logical structure notation developed in this chapter has been used in place of the verb notation used in the original. ‘{. . .}’ represents qualia which are not specified for the example. The logical structure for begin would be BECOME do (x, y), where the logical structure of the complement verb fills the y variable slot.

51

52


role in the qualia for the ‘Dana began to read a novel’ interpretation and in the agentive role in the qualia for the ‘Dana began to write a novel’ reading. Since qualia express important semantic properties of nominals, qualia may be used to represent the selectional restrictions of verbs. That is, if a verb requires a particular type of argument, that argument position can be annotated with the appropriate qualia type, and then only NPs whose head noun has qualia compatible with it may function in the logical structure. An integral part of an NP is the set of nominal operators, which were introduced in section 1.6. They may be given a representation in the semantic representation of the NP analogous to the representation of operators in the semantic representation of the clause. The NP operators are summarized in (2.31). (2.31)

DEIC PROX DEF + NEG Ø QNT ∃ NUM SG NASP COUNT N

The logical structure for the scarf would be as in (2.32). (2.32)

DEF + NEG Ø QNT ∃ NUM SG NASP COUNT (scarf)

Possessive NP constructions involve a possessive predication within the NP. Possessive predications are based on have (x, y), e.g. have (woman, book) for The woman has a book, and the corresponding alienable possessive NP the woman’s book would be represented as have (woman, book), ------ with the head underlined. In a possessive predication, the first argument of have is the possessor and the second argument the possessed (see Table 2.4 below), and therefore within the NP the possessed is normally selected as the head of the NP. It is possible, however, to choose the possessor as the head, i.e. have (woman, ---------- book), yielding the woman with the book. Certain types of NP adjuncts receive a similar representation, e.g. the NP the table in the library would have the representation table). Inalienable possession is represented by have.as.part (x, be-in (library, ------y), e.g. have.as.part (woman, arm) ----- for the woman’s arm or the arm of the woman. Kin possession is expressed by have.as.kin (x, y), as in have.as.kin (woman, sister) for the woman’s sister or the sister of the woman. ------Pronouns and reflexive pronouns are represented directly in the logical structure in which they occur, as illustrated in (2.33). (2.33) a. a . b. b .

He saw Pat. see (3sgM, Pat) Mary saw herself. see (Mary, herself)

The conditions on possible intrasentential coreference involving pronouns are related to information structure; see Van Valin and LaPolla (1997), section 5.6. The conditions on the binding of reflexive pronouns like herself will be discussed in sections 5.3 and 7.5 below. Since every argument in a logical structure would receive a representation like the one in (2.32) in a full semantic representation, such representations

2.4 Semantic roles

〈IF INT 〈TNS PRES 〈ASP PERF PROG 〈do′ (x, [cry′ (x)])〉〉〉〉 〈DEF + 〈QNT

〈NUM SG 〈NASP COUNT 〈 be′ (man (x), [tall′ ])〉〉〉〉〉

Figure 2.1 Semantic representation of Has the tall man been crying?

would be extremely complex and cumbersome. An example of the representation of a simple intransitive clause, Has the tall man been crying?, is given in Figure 2.1.

2.4

Semantic roles

The next step in the development of the semantic representation is the specification of the semantic relations that obtain between a verb or other predicator and its arguments. Semantic roles have been discussed at three distinct levels of generality. The first are what may be called ‘verb-specific’ semantic roles, e.g. runner, killer, hearer, broken, etc. The second are thematic relations, which are generalizations across the verb-specific roles, e.g. agent, instrument, experiencer, theme, patient. The third are generalized semantic roles, the semantic macroroles, actor and undergoer, which are generalizations across thematic relations. Actor is a generalization across agent, experiencer, instrument and other roles, while undergoer is a generalization subsuming patient, theme, recipient and other roles. Agent is the prototype for actor, and patient is the prototype for undergoer. The relationships among the three types of semantic roles are summarized in Figure 2.2 on p. 54. RRG posits two types of semantic roles: thematic relations and semantic macroroles, which play a crucial role in the linking system.

2.4.1

Thematic relations

The logical structures introduced in section 2.1 form the heart of the lexical entry for a verb, and they correspond to the thematic relations / θ-role lists that other theories associate with a verb in its lexical entry. There is, however, no listing of thematic relations in a verb’s lexical entry in RRG; rather, thematic relations are defined in terms of the argument positions in the decomposed logical structure representations, following Jackendoff (1976). The definitions are given in Table 2.4 on p. 55. (Verbs of directed perception are listen to, whose logical structure is given in Table 2.4, look at or watch, feel, taste and smell, the last three having the same form as the corresponding stative perception verbs.) In terms of these definitions, in (2.17a) Pat is an identified, the first argument of an identificational predication, while the cup is a patient, the single argument of a one-place stative predicate of state or condition. In the locational example, the

53

54


Thematic relations Verb-specific semantic roles Giver Runner Agent Killer Speaker Dancer Thinker Believer Cognizer Knower Presumer Hearer Smeller Perceiver Experiencer Feeler Taster Liker Emoter Lover Hater Given to Recipient Sent to Handed to Seen Stimulus Heard Liked Located Theme Moved Given Broken Patient Destroyed Killed

Semantic macroroles

Grammatical relations

Actor

Subject

Undergoer

Increasing generalization, increasing neutralization of semantic contrasts

Figure 2.2 Continuum from verb-specific semantic roles to grammatical relations

library is a location and Kim is a theme.16 In the last example with see, a twoplace stative perception verb, Dana is the first argument and therefore a perceiver, while the picture is the second argument and a stimulus. The thematic relations assignments of achievement, accomplishment and semelfactive verbs are the same as the corresponding state or activity, as the addition of the operator BECOME, SEML or INGR does not affect the argument structure of the logical structure. Equational predicates are unique in that their two argument positions define the same semantic role. These predications involve only referential identity, not semantic roles in the usual sense. There is no semantic contrast between (2.34a) and (2.34b). (2.34) a. b. 16

Kim’s sister is Sandy’s lawyer. Sandy’s lawyer is Kim’s sister.

The order of arguments in the semantics reflects their thematic relation and is not correlated with the word order of English or any other language. All locational predicates in every language have this logical structure, with the first argument a location and the second one a theme, and this is true regardless of the word order of the language. The relationship between the order of arguments in logical structures and the overt order of NPs and PPs in a sentence is mediated by the linking algorithm to be introduced in chapter 5.

2.4 Semantic roles

Table 2.4 Definitions of thematic relations in terms of logical structure argument positions I. STATE VERBS A. Single argument 1. State or condition 2. Existence B. Two arguments 1. Pure location

broken (x) exist (x)

x = pat ient x = entity

be-LOC (x, y)

2. Perception

hear (x, y)

3. Cognition

know (x, y)

4. Desire

want (x, y)

5. Propositional Attitude

consider (x, y)

6. Possession

have (x, y)

7. Internal Experience

feel (x, y)

8. Emotion

love (x, y)

9. Attributive

be (x, [pred ])

10. Identificational

be (x, [pred ])

11. Specificational

be (x, y)

12. Equational

equate (x, y)

x = location , y = theme x = p e r c e i v e r, y = stimulus x = cognizer , y = content x = want e r , y = desire x = judger, y = judgment x = possessor , y = possessed x = experiencer , y = sensation x = e m o t e r, y = tar ge t x = attributant , y = at t r i but e x = identified, y = identity x = var i a b l e, y = val ue x, y = r e f e r e n t

do do do do do

x = effector x = m ov e r x = s t - m ove r x = l-emitter x = s-emitter

II. ACTIVITY VERBS A. Single argument 1. Unspecified action 2. Motion 3. Static motion 4. Light emission 5. Sound emission B. One or two arguments 1. Performance

(x, Ø) (x, [walk (x)]) (x, [spin (x)]) (x, [shine (x)]) (x, [gurgle (x)])

do (x, [sing (x, (y))])

2. Consumption

do (x, [eat (x, (y))])

3. Creation 4. Directed perception

do (x, [write (x, (y))]) do (x, [hear (x, (y))])

5. Use

do (x, [use (x, y)])

x = p e r f o r m e r, y = performance x = consumer , y = consumed x = c r e at o r, y = c r e at i o n x = o b s e r v e r, y = stimulus x = u s e r, y = implement

55

56


The difference between these two is pragmatic, not semantic. Hence there is no semantic role distinction between the two arguments, and the choice of one or the other as subject appears to be driven by information structure contrasts, which will be discussed in the next chapter. Activity verbs are the other class of primitive predicates, and there are at least ten subclasses. The first argument of a non-motion activity verb is an effector, the participant which does some action and which is unmarked for volition and control. The other thematic relations associated with the first argument of activity verbs are really subtypes of effector. Activity verbs tend strongly to be singleargument, but there are some which may have two arguments, e.g. eat, drink, play. A prominent thematic relation that is missing from Table 2.4 is that of agent. While many approaches take agent to be one of the basic thematic relations, in RRG it is analysed very differently. To begin with, if agent is taken to be the intentional, volitional and controlling participant in an event, then there are many verbs which appear to take agents in some sentences but not others. (2.35) a. b. c.

The man killed his neighbour. The man intentionally killed his neighbour. The man accidentally killed his neighbour.

(2.36) a. b. c.

The man murdered his neighbour. ?The man intentionally murdered his neighbour. *The man accidentally murdered his neighbour.

(2.37) a. b.

A branch falling from Pat’s tree killed his neighbour. *A branch falling from Pat’s tree murdered his neighbour.

Murder, unlike kill, requires a perpetrator which is acting intentionally and volitionally and is in control of his or her actions. This precludes sentences like (2.36c), in which there is no intention on the part of the perpetrator to carry out the action, and like (2.37b) in which the perpetrator is an inanimate entity incapable of volition, intention and control by definition. Thus, murder requires an agent argument. Kill, on the other hand, is quite compatible with a non-intentional perpetrator, as (2.35c) shows, or with an inanimate perpetrator, as (2.37a) shows. Facts like these led Holisky (1987) and Van Valin and Wilkins (1996) to argue that the basic notion that applies to both of these verbs is effector, and agents are a special type of effector. With verbs like murder, the agency is lexicalized in the meaning of the verb and therefore would have to be represented in its logical structure. Following Ross (1972), DO signals agency in logical structures, e.g. DO (x, [do (x, Ø)] CAUSE [BECOME dead (y)]) for the logical structure for murder. With verbs like kill, on the other hand, the effector argument is interpreted as an agent if its referent is human (or sentient) and if there is no information in the clause to the contrary, e.g. an adverb like accidentally. Hence the logical structure for kill would be [do (x, Ø)] CAUSE [BECOME dead (y)]. Languages differ with respect to the extent to which agency is lexicalized in verbs: English

2.4 Semantic roles

appears to have only relatively few verbs which absolutely require an agentive interpretation of their effector, whereas Japanese seems to lexicalize agency with many more verbs than English (Hasegawa 1996). Causative verbs have a logical structure composed of (typically) an activity logical structure plus a second logical structure of any of the six Aktionsart types linked by the connective CAUSE. Accordingly, the role assignments of the causative logical structure are those of the constituent activity and other logical structures; no new roles are added. Hence in (2.17g) the thematic relations of the arguments are a function of the assignments in (2.17a)–(2.17f). The representation of arguments in the logical structure of a verb or predicator provides a strict definition of ‘core argument’. In section 1.1 it was stated simply that the core contains the arguments of the predicate in the nucleus, and it is now possible to state this more precisely. All arguments which appear in the core of a simple clause must be linked to argument positions in the logical structure of the predicate in the nucleus, and, in the default situation, all arguments in the logical structure of the predicate must appear in the core of the clause. However, it is not always the case that an argument in the logical structure occurs in the core; in a passive construction, for example, the effector, if overt, will be realized as an oblique constituent in the periphery. Among core arguments a further distinction is made between direct and oblique core arguments, as noted in section 1.1. This contrast is based on the morphological coding of the arguments: direct core arguments are those that are morphologically unmarked or coded with a direct case, as in dependent-marking languages like English and German, or are crossreferenced on the verb, as in head-marking languages like Lakhota and Tzotzil. Oblique core arguments are those marked by an adposition or by an oblique case. Thus English give has three core arguments; in Dana gave Pat the key there are three direct core arguments, while in Dana gave the key to Pat there are two direct core arguments and one oblique core argument. This will be discussed in more detail in sections 4.4–4.5. It might appear from Table 2.4 that RRG posits a great many thematic relations, but in fact there are only five relevant distinctions. The five distinctions correspond to the five possible argument positions in logical structures. This may be represented as in Figure 2.3 on p. 58. Agents are wilful, controlling, instigating participants in states of affairs, while patients are strongly affected participants. Taking these as endpoints on the continuum makes it possible to place the other role-types with respect to them. The DO of lexicalized agency always co-occurs with the do (x, . . . which defines effector and its subtypes, and accordingly the first two columns are closely related to each other; all of them express participants which do something. At the other end of the continuum fall patient and theme, etc. The single argument of state predicate (x) includes those participants which are crushed, killed, smashed, shattered, broken, destroyed, etc., while the second argument of predicate (x, y) includes those participants which are placed, moved, thrown, given, possessed, transferred, seen, heard, loved, etc. In terms of affectedness, the former type of participant is much more affected than the latter,

57

58


Arg. of DO

1st arg. of do′ (x, ...

1st. arg. of pred′ (x,y)

2nd arg. of pred′ (x,y)

AGENT

EFFECTOR MOVER ST-MOVER L-EMITTER S-EMITTER PERFORMER CONSUMER CREATOR OBSERVER USER

LOCATION PERCEIVER COGNIZER WANTER JUDGER POSSESSOR EXPERIENCER EMOTER ATTRIBUTANT IDENTIFIED VARIABLE

THEME PATIENT STIMULUS ENTITY CONTENT DESIRE JUDGMENT POSSESSED SENSATION TARGET ATTRIBUTE IDENTITY VALUE PERFORMANCE CONSUMED CREATION IMPLEMENT

Arg. of state pred′ (x)

Figure 2.3 Thematic relations continuum in terms of logical structure argument positions

hence the placement of the single argument of state predicate (x) at the end of the hierarchy. Into the middle of the continuum falls the first argument of predicate (x, y). If it is contrasted with the first argument of do , it is clear that seeing, thinking, believing, possessing, etc., are less agent-like than are speaking, doing, moving, performing, consuming, hence their placement to the right of effector, etc. If, on the other hand, the contrast is with the second argument of predicate (x, y), then the reverse conclusion follows. Seeing, thinking, liking, believing, etc., involve some kind of internal activity (mental, emotional or perceptual) on the part of the participant, whereas being seen, being thought about, being liked or being believed does not require any action or effort of any kind on the part of the participant. Hence the participant denoted by the first argument is more active and hence more agent-like than the participant referred to by the second argument, and, accordingly, the first argument is closer to the agent end of the hierarchy than the second argument. Thus, the positioning of the different argument positions in the continuum in Figure 2.3 reflects the semantic contrasts among them. In addition to static locations, e.g. be-at (x, y) and possession, e.g. have (x, y), there are also change of location and possession arguments. ‘Goal’ may be defined as the location argument in the following logical structure configuration: . . . INGR/BECOME be-at/in/on (x, y). ‘Recipient’ may be defined as the possessor argument in the following configuration: . . . INGR/BECOME have (x, y). ‘Source’ may be defined as the location or possessor argument in either of the configurations: . . . INGR/BECOME NOT be-at (x, y) or . . . INGR/BECOME NOT have (x, y). The thematic relation of instrument is also missing from Table 2.4, because there are actually two distinct roles that it subsumes. The first is analysed as a

2.4 Semantic roles

type of effector, specifically as a manipulated inanimate effector in a causal chain. This is illustrated in (2.38). (2.38) a. b.

Leslie shattered the window with a rock. [do (Leslie, Ø)] CAUSE [[do (rock, Ø)] CAUSE [INGR shattered (window)]]

Note that the second part of the logical structure can be a complete clause, The rock shattered the window.17 It is of course also possible to leave out the intermediate effector, yielding Leslie shattered the window. The term ‘instrument’ will be used to refer to intermediate inanimate effectors in a causal chain. The second type is not part of a causal chain; it will be referred to as an ‘implement’. (‘∧’ is a connective meaning ‘and simultaneously’.) (2.39) a. b.

Chris ate the soup with a spoon. do (Chris, [eat (Chris, soup) ∧ use (Chris, spoon)])

Unlike in (2.38), the implement cannot function as the subject if the effector is omitted, i.e. *The spoon ate the soup. Implements can only be added to an activity logical structure. Logical structures may be expanded primarily in three ways: (1) to specify a full causal chain, as in (2.38), (2) to add an implement as in (2.39), and (3) to specify the source, path (or means) and goal (or recipient) with motion active accomplishment predications and transfer predications. The theoretical implications of this system for deriving thematic relations from logical structures are very important. If it is the case that the thematic relations which a verb takes are a function of the argument positions in its logical structure, and there is a system of lexical representation in which there are independent criteria for assigning logical structures to verbs, then there are independent criteria for assigning thematic relations to verbs. This is the case because the thematic relations are a function of the logical structure of a verb, and there are independent criteria for attributing a logical structure to a verb. Thematic relations cannot be assigned on an arbitrary basis, because logical structures cannot be assigned arbitrarily; rather, logical structures are determined on the basis of the tests in Table 2.2. Thus the great advantage to this system of lexical representation is that there are tests which provide independent criteria for assigning a particular logical structure and hence a particular argument structure to a given verb. This system also has important implications for how one actually goes about analysing a language. In order to determine the argument structure of a verb, it is first necessary to ascertain its Aktionsart in the construction in which it occurs, using the tests in Table 2.2. Having established that, its logical structure can be created, following Table 2.3, and its argument structure follows from Table 2.4. Thus, it is necessary to ascertain the Aktionsart of the verb in the sentence, and from this its argument structure follows. What is not appropriate 17

For detailed discussion of effectors and instruments, see Van Valin and Wilkins (1996) and Van Valin and LaPolla (1997), section 3.2.3.2.

59

60


in this system is to decide arbitrarily what thematic relations a verb should have and then to construct a logical structure which would yield those roles, or to start from a semantic paraphrase and then construct a logical structure to express that paraphrase. The logical structures are determined by the tests in Table 2.2. It is important to emphasize that in the system presented here, thematic relations play no direct role in lexical representation; the relevant semantic properties of the verbs are expressed by the decompositional logical structure representations, not by thematic relations. Thus even though a large number of role labels like agent, cognizer, theme and patient have been used in this discussion, they are merely mnemonics for argument positions in logical structure. They have no independent status. Since there is as yet no adequate decompositional representation for the primitive state and activity predicates which are the argument-bearing components of the system and which carry the substantive semantic load, these labels are useful in that they indicate the subclass of the predicate; hence cognizer means ‘first argument of a two-place state predicate of cognition’, judgment means ‘second argument of a two-place state predicate of propositional attitude’, and theme means ‘second argument of a two-place state predicate of location’, for example. These labels will be used in this way, and it must be kept clearly in mind that these labels do not refer to independently meaningful relations but rather to argument positions in the logical structure of predicates of a certain type. 2.4.2

Macroroles

The second type of semantic roles is generalized semantic roles, the two macroroles ‘actor’ and ‘undergoer’.18 These are the two primary arguments of a transitive predication, either one of which may be the single argument of an intransitive verb. They correspond to what is pretheoretically commonly called ‘logical subject’ and ‘logical object’, but these labels are not utilized in RRG because ‘subject’ and ‘object’ are normally used to refer to syntactic, not semantic relations. Generally speaking, the actor is the most agent-like argument, while the undergoer is the most patient-like. They are called ‘macroroles’ because each subsumes a number of specific thematic relations. Macroroles are motivated by the fact that in grammatical constructions groups of thematic relations are treated alike. For example, themes and patients function alike for certain purposes in the grammar. It is necessary to distinguish them on semantic and other grounds. But nevertheless, the grammar, for certain purposes, treats these roles as essentially the same, e.g. they can be both the direct object in an active and the subject in a passive. In fact, active and passive in English can be described in terms of lists of thematic relations. Agent, effector, experiencer, perceiver, possessor, judger, etc., can be the subject of an active verb, while patient, theme, stimulus, possessed, location, etc., can be direct object. In the English passive, patient, theme, stimulus, possessed, location, etc., can be subject, while agent, 18

See Van Valin (1999a) for a comparison of different theories of generalized semantic roles.

2.4 Semantic roles

effector, experiencer, perceiver, possessor, judger, etc., can be the object of the preposition by. It appears that a significant generalization is being missed here, since there are long disjunctive lists of roles in these statements. But in fact, it is not an accident that they seem to group together the way they do, and the obvious generalization can be captured in terms of semantic macroroles: in an active clause, the actor is subject and the undergoer direct object, while in a passive the undergoer is subject and the actor is in a peripheral PP. The relation between macroroles and logical structure argument positions is captured in the actor-undergoer hierarchy in Figure 2.4. This double hierarchy ACTOR Arg. of DO [

UNDERGOER 1st arg. of 1st arg. of do′ (x,... pred′ (x, y)

2nd arg. of pred′ (x, y)

Arg. of pred′ (x)

′ = increasing markedness of realization of argument as macrorole]

Figure 2.4 Actor–undergoer hierarchy (preliminary)

says simply that given the logical structure of a transitive verb, the leftmost argument will be the actor and that the rightmost argument will be the undergoer. This is the default situation. Marked assignments to undergoer are possible, as in the ‘dative shift’ alternation in which the first argument of the two-place state predicate, not the second, is undergoer, as in (2.40c). (2.40) a. b. c.

[do (Pat, Ø)] CAUSE [BECOME have (Chris, book)] Pat [actor] gave the book [undergoer] to Chris. Pat [actor] gave Chris [undergoer] the book.

There are languages which have only the first possibility, some which have only the second, and some which have both. The issues raised by three-place predicates and their implications for the hierarchy will be discussed further in section 4.4. The hierarchy in Figure 2.4 represents thematic relations along the continuum presented in Figure 2.3. The prototypical actor is an agent, the prototypical undergoer a patient, but effectors and arguments bearing other thematic relations can also function as actor, and arguments bearing thematic relations other than patient can also serve as undergoer; this depends on the logical structure of the particular verb, as shown in Table 2.4. It must be emphasized that the label ‘undergoer’ should not be taken literally, just as ‘actor’ should not. The actor of see does not do anything but is nevertheless an actor in the sense intended here, i.e. the logical subject; one could say that the actor is the participant which is responsible for the state of affairs, in the sense that it is impossible to have an action without an entity doing the action, a perceptual situation without a perceiving entity, or a cognitive or emotional situation without a participant experiencing the cognitive or emotional state. Similarly, the undergoer of see does not undergo anything, unlike the undergoer of e.g. kill, but it is still the undergoer of the verb, i.e. the logical object. In general, the undergoer represents the non-instigating, affected

61

62


participant in a state of affairs. The specific semantic content of the macrorole with a particular verb is supplied by the position of the argument in the logical structure, not by its macrorole status, although the two are clearly related. The ranking of argument positions (thematic relations) in Figure 2.4 with respect to actor and undergoer selection is supported by considerable crosslinguistic evidence. The maximal unmarkedness of agent as actor and patient as undergoer follows from the fact that if a verb has an agent argument, it will always be actor, and likewise if one has a patient argument, it will always be undergoer. With respect to the actor end of the hierarchy, if a verb has both a potential agent and an inanimate effector (i.e. instrument), as in (2.38), the potential agent must be the actor, never the effector, an observation that goes back at least to Fillmore’s (1968) original work on case grammar. Transitive verbs which take perceiver and stimulus arguments, e.g. see (see (x, y)), or possessor and possessed arguments, e.g. have (have (x, y)), always have the perceiver- or possessor-type as actor, never the stimulus- or possessed-type argument. With regard to the undergoer end of the hierarchy, the markedness relations are reversed from the actor end. If a verb takes perceiver- or possessor-type arguments and stimulus- or possessedtype arguments, e.g. y and z with show ([do (x, Ø)] CAUSE [BECOME see (y, z)]) and give ([do (x, Ø)] CAUSE [BECOME have (y, z)]), the z argument is the unmarked choice for undergoer, even though these verbs allow both possibilities in English. Evidence for this comes from a variety of sources. First, there are wellknown arguments for the markedness of the y-argument-as-undergoer form of the dative-shift construction in English, e.g. the restrictions on pronominal arguments in this form. Second, nominalizations provide evidence for the priority of the z argument over y: a bookshower would have to be someone who shows books to people and a studentshower could not be someone who shows things to students but rather who shows students to people; the same interpretations are found with flowergiver and girlgiver, which cannot mean someone who habitually gives things to girls. Third, in the survey of dative-shift constructions in a wide range of languages presented in Foley and Van Valin (1985), the z-as-undergoer form is clearly the unmarked form in virtually all of the languages; in most languages other than English, the base form of a verb of the appropriate class can take only the z argument as undergoer, and a derivational morpheme of some kind must be added in those languages which allow the y argument as undergoer. It is very significant that most languages do not have dative-shift constructions at all; in them, the choice of undergoer is lexically governed, and the vast majority of the relevant verbs in these languages are like English put and donate in having z not y as undergoer. Thus with respect to the undergoer end of the hierarchy, the second argument of a two-place predicate outranks the first argument of a two-place predicate for undergoer in languages like English, Russian, Dyirbal and many others. It will be shown in section 4.4, however, that there are languages which exhibit a different pattern with three-argument verbs. The number of macroroles that a verb takes is generally predictable from its logical structure; there are only three possibilities: 0, 1, 2. If a verb has two or

2.4 Semantic roles

more arguments in its logical structure, e.g. [do (x, Ø)] CAUSE [BECOME be-at (y, z)] or hear (x, y), then the unmarked situation is for it to take two macroroles. If a verb has only a single argument in its logical structure, e.g. do (x, [walk (x)]) or BECOME open (y), then the unmarked situation is for it to have only one macrorole. Verbs with no arguments, e.g. do ([snow ]), have no macroroles. The nature of the macroroles is also a function of the verb’s logical structure. If a verb takes two, then they must be actor and undergoer. For verbs which have a single macrorole, the default choice follows directly from the logical structure of the verb: if the verb has an activity predicate in its logical structure, the macrorole will be actor; otherwise, it will be undergoer. The default macrorole assignment principles are summarized in (2.41). (2.41) a.

b.

Default Macrorole Assignment Principles Number: the number of macroroles a verb takes is less than or equal to the number of arguments in its logical structure. 1. If a verb has two or more arguments in its logical structure, it will take two macroroles; 2. If a verb has one argument in its logical structure, it will take one macrorole. Nature: for verbs which take one macrorole, 1. If the verb has an activity predicate in its logical structure, the macrorole is actor. 2. If the verb has no activity predicate in its logical structure, the macrorole is undergoer.

In English, most verbs follow these defaults; the exceptions are intransitive location verbs with two arguments, e.g. lie as in The book is lying on the table, or motion active accomplishment verbs such as run in Dana ran to the house, which have only a single macrorole, an undergoer with lie and an actor with run. This is a systematic exception, and therefore it could be handled by a general principle governing the macrorole number of locative verbs. There are also verbs like seem which have a propositional argument and an experiencer in their logical structure but which do not contribute the subject or object to the clause. Since the number of macroroles is not predictable from the number of logical structure arguments with these verbs, it would have to be specified in the lexical entry of the verb, e.g. 0 for seem. This could be formalized in terms of a simple feature such as [MR α], with values [MR 0], [MR 1] and [MR 2]. The appearance of this feature in the lexical entry of a verb signals that the default principles are overridden. The identity of the macrorole would not need to be stipulated, however, since it follows from the general principle in (2.41b). One class of verbs appears to be a universal exception to the default generalization regarding macrorole number: activity verbs. Multiple-argument activity verbs with a non-referential second argument, e.g. beer in Kim drank beer, never have an undergoer macrorole. As noted above, the undergoer NP refers to the non-instigating, affected participant in the state of affairs denoted by the clause, and because beer in this example is

63

64


non-referential, it cannot refer to a specific affected entity; hence it cannot be an undergoer. In many languages this argument appears incorporated into the verb and not as an independent NP. Consequently, two-argument activity verbs with a non-referential second argument take only an actor macrorole. Those with a referential second argument, e.g. verbs of directed perception and use, do take two macroroles. Because this is a general principle of the theory, it is not necessary to put ‘[MR 1]’ in the lexical entry of verbs like eat, drink, read, etc. The active accomplishment uses of these verbs are fully transitive and take two macroroles. The macrorole number of a verb corresponds closely to the characterization of a verb in terms of the traditional notion of transitivity: single macrorole verbs are intransitive, two macrorole verbs are transitive. The traditional notion refers to the number of direct NPs that appear in the syntax, and this corresponds to the number of direct core arguments, in RRG terms. It is necessary, then, to distinguish between what Narasimhan (1998) calls ‘macrorole transitivity’ (Mtransitivity), which refers to macrorole number, and syntactic transitivity (Stransitivity), which refers to number of direct core arguments. The number of direct core arguments need not be the same as that of macroroles; there are never more than two macroroles, but in a sentence like Leslie sent Pat a card there are three direct core arguments. Also, in Pat drank beer there are two direct core arguments but only one macrorole. From an RRG perspective, the S-transitivity a verb takes is less indicative of its syntactic behavior in simple sentences than its M-transitivity, and consequently ‘transitivity’ is understood in RRG as Mtransitivity and is defined in terms of the number of macroroles a verb takes: 2 = transitive, 1 = intransitive, and 0 = atransitive. This is summarized in Table 2.5. Table 2.5 Macrorole number and transitivity

snow die drink [activity ] drink [act accompl ] kill set send

Semantic valence

Macrorole number

M-transitivity

0 1 1 or 2 2 2 3 3

0 1 1 2 2 2 2

Atransitive Intransitive Intransitive Transitive Transitive Transitive Transitive

There are many three-argument verbs but only two macroroles; why is there no third macrorole? Two possible justifications for a third macrorole are (1) labelling the third argument of a ditransitive verb, and (2) accounting for dative case assignment. These have no force in RRG. The third argument of a ditransitive verb is a non-macrorole core argument; in German, for example, it would be a non-macrorole direct core argument, since it is not adpositionally marked, while

2.4 Semantic roles

in English it would be a non-macrorole oblique core argument in (2.40b), since it is adpositionally marked, but a non-macrorole direct core argument in (2.40c), since it is not. Dative case assignment does not require the postulation of a third macrorole, as will be shown in section 4.4.19 There are strong empirical and theoretical reasons for rejecting the postulation of a third macrorole. First and foremost, it is highly likely that it would not be universal like actor and undergoer. While all languages have cores with two core arguments, some languages strongly disprefer and perhaps even do not permit three core arguments in a single core. Some serializing languages, e.g. Yoruba, Yatye (Stahlke 1970), fall into this category. In such languages, clauses with more than two arguments require complex expressions in which the additional argument is a core argument of a second nucleus in a second core. So, for example, in expressing a transfer a verb meaning ‘give’ would be serialized with the transfer verb in order to express the recipient, or with a verb like ‘break’ or ‘kill’ a verb meaning ‘take’ or ‘use’ would be serialized in order to express the instrument. Second, across languages which permit three core arguments, there is no consistent morphosyntactic treatment of the third argument. Actor and undergoer, on the other hand, do have certain consistent coding properties across languages: in active voice clauses they are always direct arguments of the verb, never oblique. What ‘direct’ means morphosyntactically varies from language to language: in English it means not being marked by a preposition; in German and Russian it means being case marked in a direct case and not marked by a preposition; in head-marking languages like Lakhota (Siouan) and Tzotzil (Mayan), it means being coded on the verb. In case-marking languages, actor and undergoer are either nominative and accusative or ergative and absolutive. By contrast, there is no consistent treatment of the third core argument (Faltz 1978): it may be a direct argument in the dative case, e.g. German, Russian, Dyirbal, or an oblique argument marked by an adposition, e.g. English, Jakaltek. Actor and undergoer are never oblique arguments within the core. This raises a further issue: what exactly would count as a third macrorole? In a language like German or Russian, for example, it could be restricted to the third direct core argument of ditransitive verbs. But in a language like English, this would imply that only the to-PP with certain verbs would count as being the third macrorole. Why should this particular argument be so analysed and not other oblique core arguments? In particular, if to Chris is the third macrorole argument in (2.40b), then why shouldn’t on the truck be given the same analysis in a sentence like Dana loaded the hay on the truck? Both PPs are omissible, and both can occur as ‘direct object’ in an alternative clause pattern, e.g. (2.40c) and Dana loaded the truck with the hay. Furthermore, if on the truck has this status in the first sentence, then shouldn’t with the hay also be analysed the same way in the second? In short, it is difficult to justify why some oblique core arguments should 19

When the full range of the use of the dative case is examined, especially its uses in complex sentences, explaining it in terms of a third macrorole would be quite impossible; see Van Valin and LaPolla (1997), section 9.2.2.

65

66


be analysed as instantiating a third macrorole but not others, but if all oblique core arguments are so analysed, then whatever function and semantic content it would have would be very different from that of the hypothesized third macrorole in German and Russian. Third, a third macrorole would be markedly less important for the syntax than actor and undergoer and hence is difficult to justify on syntactic grounds. It would play little or no role in subject selection with intransitive verbs. The single argument of an intransitive verb is either an actor or an undergoer in the vast majority of cases, and in those cases where the single argument is a non-actor or non-undergoer it does not correspond semantically to the third argument of three-argument verbs. It also plays no role in the major typology of syntactic systems: ergative vs accusative vs split-intransitive (e.g. Acehnese) (see chapter 4). These differences revolve around the treatment of actor and undergoer; the third argument of ditransitive verbs is not a factor. Thus, a third macrorole would be a qualitatively different concept from the two semantic macroroles posited in RRG. It would not be universal, it would not receive consistent morphosyntactic treatment, and it would be relatively unimportant for the syntax. There is, then, no justification for positing a third macrorole and good reasons not to postulate one. Examples of partial lexical entries for some English verbs are given in (2.42). (2.42) a. b. c. d. e. f. g. h. i. j. k. l. m. n. o.

smash receive own belong (to) arrive go appear see watch show run eat devour freeze be afraid

[do (x, Ø)] CAUSE [BECOME smashed (y)] BECOME have (x, y) have (x, y) have (x, y) [MR1] BECOME be-at (x, y) do (x, [move.away.from.ref.point (x)]) & INGR be-at (y, x) appear (x, y) [MR0] see (x, y) do (x, [see (x, y)]) [do (w, Ø)] CAUSE [BECOME see (x, y)] do (x, [run (x)]) do (x, [eat (x, y)]) do (x, [eat (x, y)]) & INGR consumed (y) BECOME frozen (x) feel (x, [afraid (y)])

No list or other explicit statement of the thematic relations associated with a verb has to be specified, since they are derived from the logical structure, and because of the principles in (2.41), nothing more is needed for indicating transitivity and macrorole choice. Only with exceptional verbs like appear and belong to must the feature [MR α], be listed. There is no need for any syntactic subcategorization information in the lexical entries of verbs in RRG. Lexical idiosyncrasies are specified in the lexical entry. For example, show in (2.42j) allows either its second or third argument to be undergoer (the well-known ‘dative alternation’; see section 4.4), whereas donate does not; hence it is necessary to specify that the

2.4 Semantic roles

third argument is the only possible choice for undergoer with donate, i.e. [do (x, Ø)] CAUSE [BECOME have (y, z)], U = z. It should be noted that the prepositions which mark the oblique core arguments of verbs like show are not stated in the lexical entries, as they are in all other theories; this is because they can be predicted by a general rule and therefore need not be listed (Foley and Van Valin 1984; Jolly 1991, 1993); see section 4.4. The macroroles of actor and undergoer function as the interface between thematic and grammatical relations. Just as actor is not equivalent to agent, it is likewise not equivalent to syntactic subject; nor is undergoer equivalent to syntactic direct object. This can be seen clearly in (2.43). (2.43) a. b. c. d. e.

Chris [SUBJ, ACTOR] drank the beer [DOBJ, UNDERGOER]. The beer [SUBJ, UNDERGOER] was drunk by Chris [ACTOR]. Chris [SUBJ, ACTOR] drank beer [DOBJ]. The fireman [SUBJ, ACTOR] ran into the burning building. The lawyer [SUBJ, UNDERGOER] became upset over the decision.

In (2.43a) the actor is subject and the undergoer direct object, while in the passive in (2.43b) the undergoer is subject and the actor is a peripheral oblique. In (2.43c) beer is not undergoer because it is non-referential, drink being an activity verb in this sentence. The subject in (2.43d) is an actor, while the subject in (2.43e) is an undergoer. The status of grammatical relations in RRG and their interaction with semantic roles is the topic of chapter 4.

67

3

Information structure

3.0


The final step in this preliminary phase of the exploration of the syntax, semantics and pragmatics interface is the characterization of the information (focus) structure of sentences. The issue of the distribution of information in clauses and sentences is one of the most important questions for linguists studying the interaction of form and function in language, and it has ramifications for all aspects of grammar. Research on this topic goes back at least to the work of Prague School linguists such as Mathesius in the 1920s.1

3.1

Topic and focus

The approach taken here builds upon Lambrecht’s (1986, 1987, 1994, 2000) theory of information structure. In his theory, Lambrecht identifies topic and focus as the two primary information statuses that referring expressions may have in an utterance. These terms will be used as labels for discoursepragmatic functions only and not for the structural positions in which they may be manifested. He adopts the definitions for topic and comment proposed in Gundel (1988). An entity, E, is the topic of a sentence, S, iff in using S the speaker intends to increase the addressee’s knowledge about, request information about, or otherwise get the addressee to act with respect to E. A predication, P, is the comment of a sentence S, iff in using S the speaker intends P to be assessed relative to the topic of S. (Gundel 1988:210)

The notion of ‘comment’ is related to that of focus, as will be seen below. Two very important points are, first, not every utterance has a topic, and second, the topic element need not be the first element in a sentence. There is a direct and fundamental relationship between the element functioning as topic and the 1

Mathesius (1928, 1929); more recent work in this area has included Halliday (1967, 1994), Jackendoff (1972), Kuno (1972a,b, 1975), Prince (1981a,b), Sgall, Hajiˇcová and Panevová (1986), Gundel (1988), Vallduv´ı (1992), Dryer (1996), Erteschik-Shir (1997), among others.

68

3.2 Focus structure and focus types

pragmatic presupposition associated with a sentence. Lambrecht characterizes it as follows: What must be presupposed in the case of a topic is not the topic itself, nor its referent, but the status of the topic referent as a possible center of interest or matter of concern in the conversation . . . [T]he topic referent is active or accessible in the discourse . . . [T]he topic is contained in the pragmatic presupposition or is an element of the pragmatic presupposition. (1986:102)

Lambrecht (1994) gives the following definitions of pragmatic presupposition and assertion: pragmatic presupposition : The set of propositions lexicogrammatically evoked in an utterance which the speaker assumes the hearer already knows or believes or is ready to take for granted at the time of speech. (52) p r agmatic assertion: The proposition expressed by a sentence which the hearer is expected to know or believe or take for granted as a result of hearing the sentence uttered. (52)

Lambrecht’s pragmatic assertion corresponds to Gundel’s notion of comment. The focus of an utterance is the part that is asserted in a declarative utterance or questioned in an interrogative utterance. focus, or focus of the assertion: The semantic component of a pragmatically structured proposition whereby the assertion differs from the presupposition. (213)

These distinctions will be exemplified in the next sections in the discussion of patterns of information distribution and their morphosyntactic realization in different languages.

3.2

Focus structure and focus types

Information distribution in sentences is of concern to syntacticians only to the extent that it has morphosyntactic expression and ramifications. The grammatical system which serves to indicate the scope of the assertion in an utterance in contrast to the pragmatic presupposition is termed the ‘focus structure’ by Lambrecht. He characterizes it as follows: focus structure : The conventional association of a focus meaning [distribution of information] with a sentence form. (1994:222)

A major component of his theory of focus structure is a taxonomy of focus types. The fundamental contrast is between broad and narrow focus; in narrow focus the focus domain extends over only a single constituent, e.g. an NP, while in broad

69

70

information structure

focus it encompasses more than one constituent. There are two kinds of broad focus, predicate focus and sentence focus. Predicate focus is universally the unmarked type and coincides with the traditionally recognized ‘topic-comment’ organization of information in a sentence. Lambrecht (2000) gives the following definition. Predicate focus structure: Sentence construction expressing a pragmatically structured proposition in which the subject is a topic (hence within the presupposition) and in which the predicate expresses new information about this topic. The focus domain is the predicate phrase (or part of it). (2000:615)

The following examples from English, Italian, French, Croatian and Japanese illustrate predicate focus constructions; the constituent bearing focal stress is in all caps. (3.1) Q: A: a. b. c. d. e.

How’s your car? My car / it broke down. (La mia macchina) si e` rotta. (Ma voiture) elle est en panne . Auto se pokvario / pokvario se. (Kuruma wa) kosyoo-si-ta.

English Italian French Croatian Japanese

In each of these examples, there is an NP functioning as topic; it is the subject NP in English, Italian and Croatian, the detached NP in French, and the wa-marked NP in Japanese. The syntactic variation across these five languages highlights the problem with the term ‘subject’ in Lambrecht’s definition; as he himself notes, the term as used here should not be construed narrowly as referring to grammatical subjects only but must include detached NPs as in French and Japanese. The information structure of the English example can be represented as follows (Lambrecht 1994:226): (3.2)

Sentence: Presupposition: Assertion: Focus: Focus domain:

My car broke d ow n . ‘Speaker’s car is available as a topic for comment x’ ‘x = broke down’ ‘broke down’ Verb plus remaining post-verbal core constituents

Sentence focus constructions differ strikingly from predicate focus constructions, in that they have no topical subject; the focus domain is the entire sentence. Lambrecht exemplifies this focus type with the following examples. (3.3) Q: A: a. b. c. d. e.

What happened? My car broke down. Mi si e` rotta la macchina . J’ai ma voiture qui est en panne . Pokvario mi se auto. / auto mi se pokvario. kuruma ga kosyoo-si-ta.

3.2 Focus structure and focus types

The entire sentence is being asserted; there is no presupposed topic, as in (3.1). The most common use of this focus type is presentational constructions, as in (3.4). (3.4) a. b. c.

Once upon a time there was an old man and a dog. Then out from under the bed ran a mouse. There arose a violent storm.

These sentences lack an established topic, and they serve to introduce new participants into the discourse. The subject NP appears in the postverbal position normally reserved for objects, the unmarked focus constituent in a predicate focus construction.2 In Italian the subject must appear postverbally in this construction, while in French it appears after a kind of ‘dummy’ verb, due to the lack of the kind of subject-inversion construction found in Italian. In Japanese, the particle marking the subject changes from wa to ga. In Croatian, the Italian and English options are both available. Lambrecht (2000) gives the following characterization of sentence focus constructions. Sentence focus structure: Sentence construction formally marked as expressing a pragmatically structured proposition in which both the subject and the predicate are in focus. The focus domain is the sentence, minus any topical non-subject arguments. (2000:617)

The analysis of the English sentence focus example in (3.3) may be laid out as in (3.5). (3.5)


My c a r broke down. None ‘Speaker’s car broke down’ ‘Speaker’s car broke down’ Clause

The final focus type is narrow focus, in which the focus domain is a single constituent; it may be subject, object, an oblique, or even the verb.3 Examples from Lambrecht are given in (3.6). (3.6) Q: A: a. b. c. d. e.

I heard your motorcycle broke down? My c ar broke down. / It was my car that broke down. Si e` rotta la mia macchina . / E` la mia macchina che si e` rotta. C’est ma voiture qui est en panne. aut o mi se pokvario. / Pokvario mi se auto. kuruma ga kosyoo-si-ta.

Here there is a definite presupposition associated with the sentence, ‘something broke down’, and the assertion is that it is the speaker’s car rather than something 2 3

See Lambrecht (2000) for a detailed discussion of the properties of sentence focus constructions cross-linguistically. Lambrecht (1994, 2000) refers to this focus type as ‘argument focus’; however, the term ‘narrow focus’, from Lambrecht (1986), will be used here, since it is possible to have narrow focus on adjuncts as well as arguments.

71

72


else. Hence the focus domain is restricted to the NP car. In English and Croatian, the subject is stressed, and in Japanese the subject and the particle ga are stressed. Italian, English and French can use cleft constructions here,4 and Italian and Croatian also have the option of using an inverted subject. (3.7)


My c a r broke down. ‘Speaker’s x broke down’ x = ‘car’ ‘car’ NP

Lambrecht (1986) distinguishes unmarked narrow focus from marked narrow focus, the difference being where the narrow focus falls: if it falls on the final constituent in the core in English, then it is unmarked, whereas if it falls to the left or to the right of that, it is marked. This is illustrated in (3.8). (3.8) a. b. c. d. e.

Leslie sent the book to dana yesterday. Leslie sent the book to Dana yesterd ay. Leslie sent the book to Dana yesterday. Leslie sent the book to Dana yesterday. leslie sent the book to Dana yesterday.

Thus narrow focus on an object is a case of unmarked narrow focus, while narrow focus on a subject is a case of marked narrow focus. Languages have different unmarked focus positions, depending largely, but not entirely, on their basic word order. In SVO languages, it is the last position in the core, e.g. as in English in (3.8), or the immediate postverbal position, e.g. in Chichewa ˆ (Bresnan and Kanerva 1989). In verb-final languages, the unmarked focus position is the immediate preverbal position (Kim 1988), e.g. Korean, Turkish. Some SOV languages make a distinction between unmarked predicate focus (i.e. verb + immediate postverbal XP) vs marked predicate focus (i.e. verb + more than one preverbal XP), e.g. Korean (Han 1999). In VOS languages, it is the immediate postverbal position, e.g. Toba Batak (Van Valin 1999b). A very common example of a narrow focus sentence is a WH-question like What did you buy? and the answer I bought ; the WH-word and the NP filling its slot in the reply are both unmarked narrow foci. In a yes–no question like Did J o h n leave? and the response No, F r e d did, John and Fred are marked narrow foci. There are further divisions that can be made in terms of types of narrow focus (e.g. Prince 1981a, Dik 1989, Bearth 1992), and a particularly significant one is between completive and contrastive narrow focus. Completive focus is the answer to a yes–no or WH-question, as in the examples above. Contrastive focus, on the other hand, involves an explicit choice among alternatives, as in (3.6). The two may co-occur in a single clause, as in (3.9).

4

See Pavey (2004) for an analysis of it-clefts in RRG.

3.3 Morphosyntactic marking: focus structure

(3.9) a. b.

Who did Bill give the book to and who did he give the magazine to? He gave the book to Mary and the m agazine to Sally.

In (3.9b), the book and the magazine are contrastive narrow foci, while Mary and Sally are completive narrow foci.

3.3

Morphosyntactic marking of focus structure

As is clear from the examples in (3.1), (3.3) and (3.6), languages employ different grammatical means for indicating the various focus constructions; syntax, morphology and prosody are all used. English makes use of both prosodic and syntactic devices. The basic mechanism is accentuation, with the main sentence stress falling on the primary focal element, as noted in section 3.2 (see also Erteschik-Shir and Lappin 1983; Selkirk 1984, 1995). English allows the focal stress to fall on any constituent in a sentence, and consequently the contrasts among the three focus types can be signalled solely prosodically. However, there are syntactic ways to mark them as well. Inversion may play a role in the presentational sentence focus construction, as in (3.4), and an it-cleft can be used to express narrow focus, e.g. it was my car that broke down (cf. (3.6a)). Focus structure affects the type of referring expression that is selected to fill a variable position in LS, because the kind of referring expression that is chosen reflects the status of the referent in the discourse context. This is captured in Figure 3.1, which is based on proposals from Givón (1983), Levinson (1987), Ariel (1990), Gundel et al. (1993), Lambrecht (1994) and Fretheim and Gundel (1996). Zero coding is the least marked coding for a topic referent, while realization as an indefinite NP is the least marked coding for a focal referent. While indefinite NPs can be topics under special contextual circumstances, it is impossible for a focal referent to be realized as zero. Markedness of occurrence as focus Zero

Clitic/Bound pronoun

Pronoun [–Stress]

Pronoun [+Stress]

Definite NP

Indefinite NP

Markedness of occurrence as topic

Figure 3.1 Coding of referents in terms of possible functions

Japanese uses primarily morphological means, the well-known particles wa and ga, to signal the different focus types. Kuno (1973) argues that (unstressed) wa is a topic marker, which accounts for its use in predicate focus constructions, and further that there are in fact two ga particles, neutral description ga (unstressed) and exhaustive listing ga (stressed). Neutral description ga is used in sentence focus constructions like (3.3e), as well as in predicate focus constructions in certain contexts (Shimojo 1995), while exhaustive listing ga is found in narrow

73

74


focus constructions. In Huallaga Quechua (Weber 1989), there is a topic marker -qa, and evidential clitics indicate the focus of the sentence. This is illustrated in (3.10). (3.10) a.

b.

Machka papa-ta-shi qara-n wamra-ta-qa. mealy potato-ACC-EVID give-3sg child-ACC-TOP ‘She [the witch] gave [her] mealy potatoes, the child.’ Chawra machka papa-ta qara-pti-n-qa wamra-qa apta-rku-n-shi. so mealy potato-ACC give-DS-3sg-TOP child-TOP grasp-DIR-3sg-EVID ‘So when she gives [her] a mealy potato, the child, she grasps it.’

The NP in the preverbal position in (3.10a), machka papa ‘mealy potato’, bears the indirect evidence evidential clitic -shi, which along with its position, indicates that it is the focus of the sentence. In (b), the event in (a) is recapitulated as a topic-marked switch-reference clause, and the verb in the main clause, which is the focus, takes the evidential clitic. Thus morphological markers serve the same function as intonation and syntax in English in indicating the various focus constructions. Italian and French both use syntactic means to distinguish among the three types, albeit differently, and in neither language is it possible to simply shift the focal stress to the preverbal subject position with no other change in the structure of the sentence, as is possible in English. Both employ the equivalent of it-clefts for narrow focus (3.6b,c), but sentence focus constructions are handled in distinct ways. In Italian the subject occurs immediately after the verb, as in (3.3b), while in French a bi-clausal construction is used in which the NP corresponding to the inverted subject in Italian is in the postverbal object position of the first clause and the substantive predicate of the sentence is in the second clause, as in (3.3c). In neither language may a focal NP occur in preverbal position (within the core), unlike English, and such a constraint is relatively common cross-linguistically. In Mandarin Chinese (LaPolla 1990) a preverbal referential NP must be part of the pragmatic presuppositions associated with the utterance. Thus a simple sentence like (3.11a) can only be interpreted as having a presupposed, specific subject; with an indefinite NP, the existential verb yˇou normally appears before the NP, rendering it postverbal, as in (3.11b). (3.11) a.

b.

Rén zài nàr. person be.at there ‘The/*a man is there.’ Yˇou rén zài nàr. exists person be.at there ‘There’s a person / someone there.’

A similar constraint holds in Kinyarwanda (Kimenyi 1980) and in the Sotho languages of southern Africa (Demuth 1989); in this group of languages, there is ‘a constraint on subjects which restricts them to being highly topical, old, given

3.3 Morphosyntactic marking: focus structure

information’ (1989:67). The following examples are from Setswana, one of the Sotho languages (Demuth 1990). (3.12) a.

b.

Monna o-bed-its-e mosimane. man 3sg-beat-PRFV-MOOD boy ‘The/*a man beat a/the boy.’ Mosimane o-bed-its-w-e ke monna. boy 3sg-beat-PRFV-PASS-MOOD by man ‘The/*a boy was beaten by a/the man,’ or ‘A man hit the boy.’

The preverbal NPs in these sentences must be interpreted as presupposed, while the postverbal NP may or may not be presupposed. This constraint has the interesting result that because WH-words are always focal in a WH-question, it is impossible for them to appear in preverbal position, as the ungrammaticality of (3.13a) shows. Consequently, in order to formulate a WH-question in which the WH-word is interpreted as a ‘subject’, it is necessary to use a passive, as in (3.13b), or a cleft construction. These examples are from Sesotho (Demuth 1989). lijo? (3.13) a. *Mang o-pheh-ile who 3sg-cook-PRFV food ‘Who cooked the food?’ b. Lijo li-pheh-li-o-e ke mang? food 3sg-cook-PRFV-PASS-MOOD by who ‘The food was cooked by who?’, ‘Who cooked the food?’

This constraint brings out clearly the fact that where the focus may fall in a sentence varies across languages; in English it may be in any position within a clause, while in Italian, French and Setswana/Sesotho it is restricted to postverbal position. Unlike the Sotho languages, WH-words may appear in the precore slot in Italian and French. This means that the constraint against focal prenuclear material applies within the core, not the clause as a whole. In Setswana, on the other hand, it holds in the clause, thereby preventing the prenuclear occurrence of WH-words. It is therefore necessary to talk about the ‘potential focus domain’ [PFD], the syntactic domain in which focus elements may occur, when discussing focus structure in a language. It contrasts with the ‘actual focus domain’ [AFD], the part of the sentence that is actually in focus. In English, the potential focus domain is the clause, whereas in these other languages the potential focus domain does not include at least some preverbal positions. This notion will become extremely important in the discussion of the focus structure of complex sentences. Some languages combine the morphological strategy of Japanese and Quechua with the syntactic strategy of French, Italian and the Bantu languages. Toura, a Mande language spoken in Ivory Coast (Bearth 1969, 1992), is just such a language. It exhibits relatively strict SOV order and has a number of means for expressing focus distinctions. The basic predicate-focus construction is exemplified in (3.14a), while two different types of focus marking are shown in (b)–(c).

75

76


(3.14) a.

T`ıa` ké gwε´ε´ lˆɔ ’. PRDM peanuts buy ‘Tia bought peanuts .’ b. Q: T`ıa` -’ mεε lˆɔ ’ le? -PRDM what buy TM ‘What did Tia buy?’ lˆɔ ’ le. A: T`ıa` -’ gwε´ε´-’ -PRDM peanuts-FOC1 buy TM ‘Tia bought peanuts.’ b . Q: Waa gwε´ε´ lˆɔ ’ le? who peanuts buy TM ‘Who bought peanuts? A: T`ıa` -’ gwε´ε´ lˆɔ ’ le. -FOC1 peanuts buy TM ‘Tia bought peanuts.’ lˆɔ ’. c. T`ıa` ké gwε´ε´-le PRDM peanuts-FOC2 buy ‘Tia bought peanuts.’

The focus marker in (b) is a tonal clitic (as is the predicate marker in many of these examples), while in (c) it is the same element that elsewhere is glossed ‘TM’. The difference between the two types of focus concerns the presuppositions involved; the type of (b) is completive focus, while that in (c) has a more contrastive function. As in English, there is a precore slot and a left-detached position in Toura; they are illustrated in (3.15). (3.15) a.

b.

Gwε´ε´-’ T`ıa` -’ lˆɔ ’ le. peanuts-FOC1 -PRDM buy TM ‘Peanuts Tia bought’, or ‘It is Peanuts (not, e.g., potatoes) that Tia bought.’ a` lˆɔ ’. Gwε´ε´ (láa` ), T`ıa` ké peanuts (TOP) PRDM 3pl buy ‘As for peanuts, Tia bought them.’

The two positions differ in Toura just as they do in English and other languages that have them; there is no intonation break between the initial NP and the following material in (a), and there is no pronoun referring to the initial NP, whereas in (b) there is an intonation break between the initial NP and the following material and there is a resumptive pronoun referring to the initial NP. The NP in the precore slot in (a) carries a focus marker, while the NP in the left-detached position in (b) carries a topic marker. Thus, Toura presents an elaborated system of focus marking which employs both special positions (precore slot, left-detached position) and focus markers for core-internal elements. This discussion raises an interesting typological point. In English, word order is very constrained and focus placement very flexible, whereas in Italian word order is very flexible and focus placement is very constrained. This contrast could be characterized in terms of how syntax and focus structure adapt to each other:

3.4 Formal representation: focus structure

in English, the focus structure adapts to the rigidity of the word order by allowing free focus placement (i.e. focus can fall on any constituent within a simple clause), whereas in Italian, the syntax adapts to the rigid focus structure (i.e. nonWH focal elements must be postnuclear) by having constructions which allow focal elements which would normally be prenuclear to occur in a postnuclear position. Hence it seems that one dimension along which languages could be characterized typologically is in terms of how syntax and focus structure interact. Van Valin (1999b) explores this interaction and finds that all four possible combinations occur: rigid syntax, flexible focus structure (e.g. English, Toura); rigid syntax, rigid focus structure (e.g. French, Toba Batak); flexible syntax, flexible focus structure (e.g. Brazilian Portuguese, Croatian, Russian, Polish (Eschenberg 1999)); and flexible syntax, rigid focus structure (e.g. Sesotho, Italian).

3.4

Formal representation of focus structure

It is necessary to integrate focus structure into the projection grammar representation of clause structure. Focus structure will be a separate projection from both the constituent and operator projections, but it is related to both. With respect to the constituent projection, predicates, arguments and peripheral PPs form the basic information units in focus structure; that is, the minimal focus domain is the nucleus, a core argument or a peripheral PP. With respect to the operator projection, the illocutionary force operator specifies the type of speech act that the sentence is in, and the potential focus domain must fall within the scope of this operator. It may be coextensive with it in simple clauses, as in English, or it may be a subset of it, as in Italian and Setswana. Within each focus structure projection both the potential focus domain and the actual focus domain will be represented. This is illustrated in Figure 3.2. The speech act node, which is related to the illocutionary force operator, anchors the focus structure projection, and the potential and actual focus domains are represented within its scope. In SENTENCE CLAUSE CORE NP

NUC

NP

PP

PRED V

Actual Focus Domain

Chris presented a child with some flowers. Basic Information Units IU IU IU IU

Potential Focus Domain

SPEECH ACT

Figure 3.2 Predicate focus in English

77

78


NP

SENTENCE

SENTENCE

CLAUSE

CLAUSE

CORE

CORE

NUC

NP

PP

PRED

PrCS

NP

NUC PRED

NP

V CHRIS

gave

IU

IU

NP

V them to her. IU

IU

SPEECH ACT

What did Chris IU

give

IU

IU

them? IU

SPEECH ACT

Figure 3.3 Narrow focus in English

these representations, the ‘IU’ nodes are the basic information units in the focus structure projection; Lambrecht (1994) argues that the minimal information unit corresponds to the minimal phrasal category in syntax. This notion will be of particular importance for the investigation of information structure in complex sentences in Chapter 6. Figure 3.2 is an example of predicate focus in English; the potential focus domain is the whole clause, and the actual focus domain is the nucleus plus the post-nuclear arguments. In Figure 3.3 two examples of narrow focus in English are given. It is possible to have two foci in a single clause, as shown in (3.9). In such a sentence there are two actual focus domains within the potential focus domain. This may be represented as in Figure 3.4.

Figure 3.4 Two types of narrow focus in a single clause, as in (3.9)

3.4 Formal representation: focus structure

Mandarin, like Italian and French, differs from English with respect to the potential focus domain in simple sentences, as noted above; in Figure 3.5, the structures of (3.16a) (predicate focus) and (3.16b) (sentence focus) are given. (3.16) a.

b.

Ch¯e lái le. vehicle come PRFV ‘The car is here.’ Lái ch¯e le. come vehicle PRFV ‘There is a car coming.’

Unlike English, the potential focus domain is not coextensive with the clause in Mandarin, and the prenuclear element in (3.16a) is outside the potential focus domain. SENTENCE

SENTENCE

CLAUSE

CLAUSE

CORE

CORE NP

NUC

NUC

PRED

PRED

V

V

Che-

lái le.

Lai

che- le.

IU

IU

IU

IU

NP

SPEECH ACT

SPEECH ACT

Figure 3.5 Predicate focus and sentence focus in Mandarin

All three projections may be combined into a single representation, as in Figure 3.6 on the next page. Each of these projections depicts information that is simultaneously present in the clause. In addition to representing focus constructions in the layered structure of the clause, the activation level of the referents of the NPs filling the argument positions in logical structures will also be indicated. For simplicity’s sake, only five levels of activation will be coded: active, i.e. actively under consideration in the discourse by means of direct mention; accessible, i.e. not actively under consideration but readily recognized by the addressee due either to knowledge of the world or to occurrence in the immediate environment of the speech situation; inactive, i.e. previously mentioned but not actively under consideration and not assumed by the speaker to be recognized by the addressee; brand new – anchored, i.e. not previously mentioned but related to something already mentioned or accessible; and brand new – unanchored, i.e. not previously mentioned or related to anything previously mentioned (Prince 1981b; Chafe 1987). Propositions may also have different levels of activation (Dryer 1996). In the dialogues in (3.1), for example,

79

80


Figure 3.6 Clause structure with constituent, operator and focus structure projections

the speaker’s car is made active by the initial question and is treated as such in the responses to it. In (3.2) the speaker’s car is inactive, as it has not been mentioned in the previous discourse. In (3.6), the proposition ‘speaker’s x broke down’ is activated, due to the question. An example of an accessible referent would be Bill Clinton, since a speaker of English could reasonably assume that their interlocutors would know who he is without him being previously mentioned in the immediate context. The different activation statuses will be indicated in LSs as in the following example, which assumes that the mayor of the city is already a topic referent established in the context. (3.17) a. b.

The mayor sent Bill Clinton a letter. [do (mayorACV , Ø)] CAUSE [BECOME have (Bill ClintonACS , letterBNU )]

The NP the mayor is activated, because its referent has already been mentioned, while the NP Bill Clinton is accessible, because its referent has not been mentioned but is assumed to be available to the interlocutors. Finally, the NP a letter is brand new – unanchored, because its referent has not been mentioned and is not assumed to be related to anything mentioned or available to the interlocutors. It will be shown in chapter 5 that different aspects of the linking algorithm are sensitive to the activation status of the referents of the NPs in logical structures.

3.5

Focus structure and the notion of VP

It was noted in section 1.2 that there is no analogue in the layered structure of the clause to the VP grouping which is basic to X-bar analyses of clause structure. However, a glance at the predicate focus constructions in

3.6 Focus structure and quantifier scope

Figures 3.2 and 3.5 reveals that the actual focus domain in these constructions corresponds to what would be a VP in an X-bar analysis. Moreover, narrow focus constructions with subject focus, as in the English examples in Figure 3.3, also isolate a VP-like grouping. This is no accident. VPs, to the extent that they exist in languages, are the grammaticalization of focus structure; they are not primitive categories in clause structure. All languages have predicate and narrow focus constructions, but the same cannot be said for VPs. (Cf., e.g., Van Valin 1987; Mohanan 1982.) Lambrecht (2000) also shows that there is no evidence for an NP–VP bipartite structure in sentence focus constructions crosslinguistically. The clearest syntactic evidence for a VP-like category in English comes from imperatives, ‘VP fronting’, ‘VP deletion’ and ‘VP anaphora’, as in (3.18). (3.18) a. b. c. d.

Open the door! I expected to find someone mowing the lawn, and mowing the lawn was Dana. Chris is having a tall latte, and Pat is, too. Q: Who mowed the lawn? A: Dana did.

In an imperative the subject is omitted, leaving the verb plus its object, a VPlike grouping. In information structure terms, unless there is contrastive stress on one of the elements, the whole utterance is in the actual focus domain of the utterance, and it corresponds to the actual focus domain of a predicate focus construction. In the other constructions, the ‘VP’ is overtly established in the discourse as topical; in (3.18b) the ‘VP’ is fronted in a sentence-focus presentational construction, while in (3.18c) it is omitted in the second clause. Example (3.18d) is a narrow focus construction with a focal subject and topical ‘VP’ (cf. Figure 3.3). In order to describe these constructions in RRG, the postulation of a VP-like category in the layered structure of the clause is unnecessary; the focus structure projection imposes these groupings on the constituent projection, and these constructions are sensitive to the pragmatically motivated bracketings of the syntactic structure. The projection grammar representation of the layered structure of the clause plus focus structure provides the basis for an explanatory representation.5

3.6

Focus structure and the interpretation of quantifier scope

The issue of the interpretation of sentences containing multiple quantifiers has long been an important topic in the study of syntax and semantics. Consider the following well-known example from English.

5

See section 7.1 for an account of ‘VP’ ellipsis which does not use the notion of VP.

81

82


(3.19) a. b.

Every girl kissed a boy.

c.

(i) Each girl kissed a different boy (‘for each girl there is a boy such that the girl kissed the boy’, i.e. [∀x, ∃y (kiss x,y), where x = girl and y = boy]) (ii) Each girl kissed the same boy (‘there is a boy such that for each girl, the girl kissed the boy’, i.e. [∃y, ∀x (kiss x,y), where x = girl and y = boy]) A boy was kissed by every girl. (=(bii), (bi))

Sentence (3.19a) is ambiguous in English, and the two interpretations are given in (3.19b). A number of linguists (e.g. Sgall, Hajiˇcová and Panevová 1986; Kuno 1991; Kuno et al. 1999; Van Valin and LaPolla 1997) have proposed that focus structure strongly affects the interpretation of quantifiers. The principle they propose can be formulated as in (3.20). (3.20) a. b.

Principle affecting quantifier scope interpretation: topical Q > focal Q Every girl kissed a boy . =(3.19b(i)) e very girl kissed a boy. =(3.19b(ii))

This principle states that (everything else being equal) a topical quantifier will have wide scope over a focal quantifier. This predicts that in a sentence like (3.19a), the default interpretation should correlate with the default focus structure, predicate focus, depicted in (3.20a). Since the universal quantifier is topical and the existential quantifier focal in (3.20a), the preferred reading should be that in (3.19b(i)), which is correct. In order to get the secondary reading in (3.19b(ii)), it is necessary to interpret (3.19a) as if it were a narrow focus construction, as in (3.20b). In this case, the existential quantifier is topical and the universal quantifier focal, and consequently the former will have wide scope, yielding the interpretation in (3.19b(ii)). This contrast is represented in Figure 3.7. SENTENCE

SENTENCE

CLAUSE

CLAUSE

CORE

CORE

NUC PRED

NP

NP

NP

KISSED

IU

IU

SPEECH ACT

NP

PRED

V Every girl

NUC V

A BOY.

IU

EVERY GIRL kissed

IU

IU

a boy. IU

SPEECH ACT

Figure 3.7 Predicate focus vs narrow focus in the interpretation of quantifiers


While this analysis can account for (3.19), is there any reason to prefer it over a purely syntactic analysis? Does it make any significant predictions? The principle in (3.20), together with the theory of focus structure in section 3.2, makes interesting predictions about the interpretation of quantifiers in Italian, Japanese and Mandarin. Consider the Italian equivalents of (3.19a,c) in (3.21), taken from Melinger (1996). (3.21) a.

b.

Ogni ragazza ha baciato un ragazzo. Every girl has kissed a boy ‘Every girl kissed a boy.’ =(3.19b(i)), =(3.19b(ii)) Un ragazzo e` stato baciato da ogni ragazza. a boy is been kissed by every girl ‘A boy was kissed by every girl.’ =(3.19b(ii)), = (3.19b(i))

The sentence in (3.21a) is the Italian translation of (3.19a), and (3.21b) is the translation of (3.19c). Unlike their English counterparts, these sentences are unambiguous: (3.21a) can only have the reading of (3.19b(i)), while (3.21b) can only have the reading of (3.19b(ii)). The lack of ambiguity is predicted by the principle in (3.20), given the analysis of the potential focus domain in Italian based on the examples in (3.3b) and (3.6b). It was pointed out in section 3.3 that in Italian, a focal core argument cannot be preverbal inside the core; in particular, a focal subject cannot occur preverbally but must be in a postverbal position. This precludes the possibility of narrow focus on a preverbal subject, which, as argued with respect to (3.20b), is what is required to derive the (3.19b(ii)) interpretation from (3.19a). Hence (3.21a) must be unambiguous, with the only possible reading being the one with the universal quantifier (ogni) having wide scope. For the same reason (3.21b) must also be unambiguous, and the only possible reading is the one with the existential quantifier (un) having wide scope. The RRG representations for (3.21a) and (3.21b) are given in Figure 3.8. The critical difference between the English structures in Figure 3.7 and the Italian ones in Figure 3.8 is the potential focus domain: it includes the subject in English but not in Italian. This difference, together with the principle in (3.20), correctly predicts the ambiguity in the English sentences and the lack of ambiguity in their Italian counterparts. Whether a Japanese subject is marked with wa vs ga has implications for the interpretation of quantifiers, following (3.20). Wa marks topics, as in (3.1e), but ga can mark both focal and topical subjects. Its use with focal subjects is illustrated in (3.3e) and (3.6e). Shimojo (1995) shows that when both the subject referent and the content of the proposition are established in discourse, i.e. neither is focal, then ga is the preferred marker for the subject. Following on the principle in (3.20), it may be predicted that the Japanese analogues of sentences like (3.19a) will be unambiguous, if the subject NP containing a quantifier is marked by wa, and will be ambiguous, if the subject NP is marked by ga. This is the case, as the sentences in (3.22) show (Watanabe 1995).

83

84


SENTENCE

SENTENCE

CLAUSE

CLAUSE

CORE

CORE
a > some (+NPsg) > every > all > most > many > several > some (+NPpl) > a few Greatest inherent tendency Least inherent tendency toward wide scope Greatest individuation, specificity Least individuation, specificity

There is a link between Ioup’s quantifier hierarchy and the principle in (3.20), because the quantifiers at the top of the hierarchy involve greater individuation and specificity of the NP; the more specific the reference of an NP is, the better it is as a potential topic, and accordingly, the quantifiers at the top of the hierarchy would yield quantified NPs which would make better topics than those at the bottom. An example of the interaction of the quantifier hierarchy with focus structure in the determination of quantifier scope can be found in the following facts from Toba Batak, an Austronesian language spoken in Indonesia (Clark 1985; W. Manik, personal communication). Toba Batak is a VOS language, and it is one of the few languages in which the basic focus structure pattern is comment– topic. The position immediately following the verb is the unmarked focus position in the clause, and positions following it, including the subject position, may be topical or focal; the default is for the subject position to be topical. In a simple sentence with two quantified NPs, the subject NP has scope over the object NP, and the resulting sentence is unambiguous. (3.25)

Manjaha buku angka guru. read book every teacher ‘Every teacher read a book.’ (= ‘every teacher read a different book’, = ‘every teacher read the same book’)

In this example both the quantifier hierarchy and the principle in (3.20) favour the wide-scope interpretation of angka guru ‘every teacher’; buku ‘book’ must be construed as non-specific. The basic interaction between the quantifier hierarchy and the pragmatic principle in (3.20) is this: when the higher ranking quantifier is also more topical, then the result should be unambiguous, but when the higher ranking quantifier is more focal and the lower ranking one is more topical, then ambiguity should result. The interesting data to be accounted for are given in (3.26)–(3.27). (3.26) a.

Mangalean missel tu tolu soridadu ganup jeneral. [Unambiguous, ganup > tolu] give missile to three soldiers each general ‘Each general is giving a missile to three soldiers.’ (I.e. ‘each general is giving a missile to a different group of three soldiers’)


b.

c. d. (3.27) a. b.

Mangalean missel tu tolu soridadu angka jeneral. [Ambiguous] ‘Every general is giving a missile to three soldiers.’ (I.e. ‘every general is giving a missile to a different group of three soldiers,’ or ‘every general is giving a missile to the same group of three soldiers’) *Diilean ganup jeneral tu tolu soridadu missel. ‘Each general gave a missile to three soldiers.’ Diilean angka jeneral tu tolu soridadu missel. ‘Every general gave a missile to three soldiers.’

[Ambiguous]

Tu tolu soridadu, mangalean missel ganup jeneral [Ambiguous] ‘To three soldiers, each general is giving a missile.’ Tu tolu soridadu, mangalean missel angka jeneral. [Unambiguous, tolu > angka] ‘To three soldiers, every general is giving a missile.’ (I.e. ‘every general is giving a missile to the same group of three soldiers’)

In (3.26a), the NP containing ganup ‘each’ is in final position, which has a default interpretation as topical; hence it has wide scope, because it is the highest ranking quantifier and is topical. In (b), on the other hand, the final NP contains angka ‘every’, and the NP containing tolu ‘three’ is not in the immediately postverbal focus position. Hence neither NP is necessarily focal, and angka is higher on the quantifier hierarchy than tolu ‘three’ but is not as high as ganup ‘each’. The default interpretation should therefore be angka ‘every’ > tolu ‘three’, but since neither NP must be interpreted as focal, it is possible to interpret tolu ‘three’ as more topical and therefore give it wide scope over angka ‘every’. In (d), the higher ranking quantifier (angka ‘every’) is in the focus position, while the lower ranking quantifier (tolu ‘three’) is in the default topic position. Each quantifier has one ‘wide-scope factor’ in its favour, and the result is scope ambiguity, as predicted. In (c), on the other hand, the highest ranking quantifier, ganup ‘each’, occurs in the focus position after the verb; the result is ungrammatical. Why should angka ‘every’ but not ganup ‘each’ be possible in this position? The answer lies in the link between the quantifier hierarchy and topicality mentioned earlier: NPs marked by ‘each’ are highly individuated and readily interpreted as topics. Hence there is a conflict between the inherently topical nature of ganup jeneral ‘each general’ and its occurrence in the focus position. The two examples in (3.27) involve a quantified NP in a left-detached, initial topic position, and both Kuno and Ioup claim that occurring in a special topic position is a strong factor in favour of wide scope. In (a) the quantifier tolu ‘three’ is in the topic position and ganup ‘each’ is in final position, also a topical position. It appears that occurring in the special topic position is such a strong factor in favour of wide scope that it balances the fact that the clause-internal quantifier is ganup ‘each’, yielding an ambiguous sentence. When the clauseinternal quantifier is lower on the quantifier hierarchy than ganup ‘each’, as in (b), the result is an unambiguous sentence, with the quantifier in the special topic position having wide scope.

87

88


It is clear, then, that information structure, the quantifier hierarchy and syntactic position interact in determining the relative scopes of quantifiers in multiple quantifier sentences in Toba Batak. Thus, the theory of focus structure presented in this chapter, the principle in (3.20), and the quantifier hierarchy together provide an explanatory account for the facts regarding quantifier scope interpretation in the English, Italian, Japanese, Mandarin and Toba Batak data presented in this section.

4

Syntactic relations and case marking

4.0


An important locus of the interaction of syntax, semantics and pragmatics is grammatical relations. RRG takes a rather different view of grammatical relations from other theories. In the first place, it does not consider them to be basic, nor does it derive them from structural configurations. Second, it recognizes only one syntactic function, not up to three like other theories; there is nothing in RRG corresponding to notions like direct object and indirect object. The syntactic function posited in RRG is not, therefore, part of the same system of oppositions as the traditional notions of grammatical relations (i.e. subject vs direct object vs indirect object), and consequently it is not really comparable to the traditional notion that is its closest analogue, subject. Third, RRG does not assume that grammatical relations are universal, in two senses. On the one hand, it does not claim that all languages must have grammatical relations in addition to semantic roles, which are universal. On the other hand, in those languages in which a nonsemantic grammatical relation can be motivated, the syntactic function posited need not have the same properties in every language; that is, the role of this syntactic function in the grammar of language X may be very different from that played by the syntactic function in language Y, and, consequently, the two cannot be considered to be exactly the same. Variation in grammatical relations systems is directly related to differences in the syntax–semantics–pragmatics interface across languages.

4.1

Do all languages have grammatical relations?

Most syntactic theories postulate that each of the core arguments bears some kind of grammatical relation, in addition to its semantic relation. The justification for positing syntactic relations in a language in addition to semantic predicate–argument relations is that there are phenomena in the language in which the distinction between two or more semantic roles is neutralized for syntactic purposes. A very simple example of this can be found in English: the verb agrees with the first NP in the core, regardless of whether it is actor or undergoer, i.e. whether it is the actor of an active-voice transitive verb, as in (4.1a), the undergoer of a passive-voice transitive verb, as in (4.1e), the actor of an intransitive verb, 89

90

syntactic relations and case marking

as in (4.1b), or the undergoer of an intransitive verb, as in (4.1c); hence the contrast between actor and undergoer is neutralized here.1 This is summarized in (4.1). (4.1) a. b. c. d. e.

The teacher has read the words. Actor of transitive V The teacher has sung. Actor of intransitive V The teacher has fainted. Undergoer of intransitive V *The teacher have read the words. *Undergoer of transitive V [active voice] The words have been read by the teacher. Undergoer of transitive V [passive voice]

The contrast between (4.1d) and (e) shows that the issue is not semantic; in both examples, the word triggering agreement is the undergoer of read. Rather, the crucial difference is the syntactic function of the NP: it is a direct object in (d) but a subject in (e). Hence the determination of the trigger for verb agreement is syntactic, not semantic. However, there is no general neutralization of the opposition among arguments bearing semantic roles; the neutralization is restricted to actor and undergoer and does not apply to every argument bearing a thematic relation to the predicate. This is, therefore, a restricted neutralization of semantic roles for syntactic purposes, and it is evidence for the existence of a syntactic predicate– argument relation in addition to the semantic relations of actor and undergoer. Most languages have phenomena such as these which motivate the postulation of grammatical relations in addition to semantic roles. Acehnese (Austronesian, Sumatra) is an example of a language in which this kind of restricted neutralization is not found, and accordingly there are no grounds for positing grammatical relations of any kind in addition to semantic roles, according to Durie (1985, 1987). In this language, statements of grammatical phenomena make reference to actor, to undergoer, or to core status; there is no construction in Acehnese which exhibits the kind of restricted neutralization of the actor–undergoer opposition of the kind described for English above. Verb ‘agreement’ (actually cross-reference, as this is a head-marking language) is sensitive to whether an argument is actor or undergoer, regardless of the transitivity of the verb; this is in sharp contrast to the situation in English, for example.2 (4.2) a.

b.

1 2 3

Gopnyan geu-mat lôn. / geu-mat-lôn.3 3sg 3A-hold 1sg / 3A-hold-1sgU ‘(S)he holds me.’

NPA A-V(-U) NPU

(Lôn) lôn-mat gopnyan. 1sg 1sgA-hold 3sg ‘I hold him/her.’

NPA A-V(-U) NPU

The same principle applies to inverted constructions like those in (4.18) and (4.19), discussed below, because the postverbal NP is still the first NP in the core. All Acehnese examples are from Durie (1985, 1987). In the clause structure schemata on the right, ‘NPA ’ = Actor NP, ‘NPU ’ = undergoer NP, ‘A-’ = actor proclitic, ‘-U’ = undergoer enclitic. The undergoer clitic cannot occur if the verb is immediately followed by the undergoer NP, unless the NP is marked by the focus particle di.

4.1 Do all languages have grammatical relations?

c.

(Gopnyan) geu-jak. / *gopnyan jak(-geuh). 3sg 3A-go go(-3U) ‘(S)he goes.’

NPA A-V / *NPU V(-U)

d.

(Lôn) lôn-jak. / *lôn jak(-lôn) 1sg 1sgA-go go(-1sgU) ‘I go.’

NPA A-V / *NPU V(-U)

e.

Gopnyan rhët(-geuh). / *gopnyan geu-rhët 3sg fall(-3U) 3A-fall ‘(S)he falls.’

NPU V(-U) / *NPA A-V

f.

Lôn rhët(-lôn) / *lôn lôn-rhët 1sg fall(-1sgU) 1sgA-fall ‘I fall.’

NPU V(-U) / *NPA A-V

The contrast between actor and undergoer in an Acehnese clause is coded primarily in the clitic pronouns on the verb: the obligatory proclitic indicates the person, number and social status of the actor (geu- codes a familiar third person argument, ji- one of higher social status), while the optional enclitic signals the same information about the undergoer. Note that the form of the clitic pronouns is the same in both instances; only position distinguishes actor from undergoer. The distinction between actor and undergoer is preserved with intransitive verbs, as (4.2c–f) show; jak ‘go’ takes only the actor proclitics, while rhët ‘fall’ takes only the optional undergoer enclitics. In other words, the verb always agrees with actors and undergoers the same way regardless of the transitivity of the verb; hence there is no neutralization of the actor–undergoer opposition in agreement (cross-reference). The actor–undergoer contrast expressed in the verbal cross-reference is maintained in the syntax. In control constructions with verbs like tém ‘want’, for example, the omitted argument in the dependent core is always the actor. This is illustrated in (4.3). (4.3) a.

b.

4

Gopnyan geu-tém [(*geu-)jak]. 3sg 3A-want (*3A-)go ‘(S)he wants to go.’

Actor of intransitive V

Geu-tém [(*geu-)taguen bu]. 3-want (*3A-)cook rice ‘She wants to cook rice.’

Actor of transitive V

c.

*Gopnyan geu-tém [rhët]. 3sg 3A-want fall ‘(S)he wants to fall.’

*Undergoer of intransitive V

d.

*Aneuk agam nyan ji-tém [geu-peuréksa lé dokto]. *Undergoer of transitive V child male that 3A-want 3A-examine by doctor ‘That child wants to be examined by the doctor.’4

The passive translation should not be taken to indicate that the Acehnese construction is a passive. Durie (1985, 1988) shows that this construction is an active sentence with a postposed actor dokto ‘doctor’; note that the actor is still cross-referenced on the verb in the usual way by a proclitic. Aneuk agam nyan ‘that boy’ is the actor of tém ‘want’ and the undergoer of peuréksa ‘examine’.

91

92


In (4.3c) the undergoer verb rhët ‘fall’ is impossible, even though its undergoer enclitic has been omitted. In the final example, the undergoer of a transitive verb has been omitted, and this too is ungrammatical. In this construction, the omitted argument must always be an actor. Thus a general notion of syntactic subject subsuming the subject of transitive verbs and the single argument of intransitive verbs is inappropriate for this Acehnese construction; rather, the restriction can simply be stated in terms of the notion of actor, without reference to transitivity or any non-semantic syntactic relation. There is a restriction but no neutralization. This construction is representative of many major constructions in Acehnese syntax, according to Durie (1985, 1987). There are also constructions in Acehnese in which the relevant restriction is stated in terms of undergoer, rather than actor (or subject). Possessor ascension is illustrated in (4.4). (4.4) a.

b.

c.

d.

e.

Seunang até lôn. happy liver 1sg ‘I am happy.’ (lit: ‘My liver is happy.’) Lôn seunang-até. 1sg happy-liver ‘I am happy.’ Ka lôn-tët rumoh gopnyan. IN 1sgA-burn house 3sg ‘I burned her house.’

Undergoer of intransitive V

Gopnyan ka lôn-tët-rumoh. Undergoer of transitive V 3sg IN 1sgA-burn-house ‘I burned her house,’ or ‘She had her house burned by me.’ *Gopnyan ka aneuk-woe. 3sg IN child-return ‘His/her child returned.’

*Actor of intransitive V

The occurrence of a possessor NP (lôn in (4.4a,b)) outside of the possessive NP (até lôn ‘my liver’ in (4.4a)) is possible only if the possessive NP functions as as the undergoer of the clause; as (4.4e) shows, this is not possible when the possessive NP serves as actor. All of the Acehnese phenomena discussed so far involve restrictions (actoronly or undergoer-only) but no neutralizations. Several major constructions are sensitive only to whether the argument in question is a core argument, e.g. NP fronting, raising, and relativization; it makes no difference whether the argument is actor, undergoer or what Durie labels a ‘dative argument’. Here the actor– undergoer opposition is indeed neutralized, but there is no restriction. Hence it is not a restricted neutralization, unlike the one from English regarding verb agreement discussed above; ‘any core argument’ is not a grammatical relation. That is, the English neutralization nullifies the actor–undergoer contrast, and it applies only to macrorole arguments; it is not a general neutralization, in that macrorole arguments are still singled out for special treatment in this construction.

4.1 Do all languages have grammatical relations?

In Acehnese, on the other hand, this neutralization applies in all clauses regardless of the transitivity of the verb, and non-macrorole arguments may be involved in the construction, as illustrated in (4.5) with raising. It is, therefore, a general neutralization of the opposition among core arguments. (4.5) a.

Gopnyan teuntèe [geu-woe]. 3sg certain 3A-return ‘(S)he is certain to return.’


b.

Gopnyan teuntèe [meungang-geuh]. 3sg certain win-3U ‘(S)he is certain to win.’


c.

Gopnyan teuntèe [geu-beuet hikayat prang sabi]. 3sg certain 3A-recite epic ‘He is certain to recite the Prang Sabi epic.’

Actor of transitive V

d.

Hikayat prang sabi teuntèe [geu-beuet]. epic certain 3A-recite ‘The Prang Sabi epic is certain to be recited by him.’

Undergoer of transitive V

e.

Gopnyan lôn-anggap [na neu-bi pèng baroe]. ‘Dative’ of ditransitive V 3sg 1sg-consider BE 2A-give money yesterday ‘I believe him to have been given money by you yesterday.’ (lit.: ‘I consider himi [you gave money [to] —i yesterday].’)

An argument of the verb in the linked core appears in clause-initial position: it is the actor of an intransitive verb in (4.5a), the undergoer of an intransitive verb in (4.5b), the actor of a transitive verb in (4.5c), the undergoer of a transitive verb in (4.5d), and the dative argument of a three-argument verb in (4.5e). Thus the neutralization found in these Acehnese examples is not like the kind found in English and most other languages and does not support the postulation of grammatical relations like subject and direct object, because it is an unrestricted neutralization. Rather, the grammar of Acehnese operates with the concepts of actor, undergoer and core argument. Acehnese is very significant for theories of predicate–argument relations, for two reasons. First, it exemplifies the rare type of language in which there is no evidence for the postulation of grammatical relations in addition to semantic predicate–argument relations; it is therefore evidence that grammatical relations are not universal in the first sense discussed in section 4.0. Second, the generalizations regarding Acehnese syntax formulated by Durie would not be possible without macroroles. If they were stated in terms of thematic relations alone, each statement would include a disjunctive list of thematic relations, e.g. ‘possessor raising is possible only if the possessive NP functions as patient, theme, experiencer or locative in its clause’. The macrorole undergoer subsumes these thematic relations and thereby makes possible an elegant statement of this restriction.

93

94

4.2


Privileged syntactic arguments

Languages like Acehnese are very unusual; the vast majority of languages do provide evidence for the postulation of grammatical relations in addition to semantic roles. The traditional description of these phenomena is in terms of the grammatical relations of subject, direct object and indirect object, but the investigation of Philippine, ergative and active languages has shown that analyses based on the traditional approach and its modern derivatives are highly problematic.5 The central concept in RRG for handling these phenomena is the ‘privileged syntactic argument of a grammatical construction’ [PSA]. It may be characterized as follows. In all languages there are syntactic constructions in which there are restrictions on the NPs and PPs (arguments and non-arguments) that can be involved in them; these restrictions define a privileged syntagmatic function with respect to that construction. In Acehnese these restrictions can be formulated in terms of either the specific semantic macroroles of actor and undergoer or the general syntactic notion of core argument. Hence the privileged syntagmatic function is semantically defined with respect to the constructions in (4.2)–(4.4). In languages like English, Icelandic and Dyirbal, on the other hand, there is a restricted neutralization of semantic roles with respect to the privileged syntagmatic function in most syntactic constructions. The restriction on arguments which can be involved in them is not characterizable in purely semantic-role terms, as in Acehnese; rather, it must be defined non-semantically, i.e. syntactically. The NP bearing the syntactically defined privileged syntagmatic function is the privileged syntactic argument of the construction. In order for a privileged syntactic argument to exist, there must be a restricted neutralization of semantic roles associated with the privileged function in the construction; if there is no restricted neutralization, as in Acehnese, then there are no grounds for positing specific non-semantic relations like subject and direct object. ‘Core argument’, while a syntactic status, does not define a specific argument relation like subject or direct object; it is neutralization without a restriction. In the English verb agreement examples in (4.1), the first NP in the core bears the defining syntactic function (it triggers verb agreement) and there is a restricted neutralization associated with it; it is therefore the privileged syntactic argument of the construction. In the construction in (4.6), known as a ‘control construction’, there is a restricted neutralization with respect to the omitted argument in the infinitival core. (4.6) a. b. c.

5

Chris wants to drink a beer. Chris wants to sing in the park. Chris wants to be stronger.

Actor of transitive V Actor of intransitive V Undergoer of intransitive V

See Van Valin (1981), Foley and Van Valin (1984, 1985), Durie (1987), Andrews (1985), Bhat (1991), Dryer (1997), among others, for detailed discussion.

4.2 Privileged syntactic arguments

d. e.

*Chrisi doesn’t want the journalist *Undergoer of transitive V [active] to interview —i . Chris doesn’t want to be interviewed Undergoer of transitive V [passive] by the journalist.

In strictly syntactic terms, there is a missing argument in the linked core in (4.6); hence in (4.6) the privileged syntagmatic function defining the construction is that of the missing argument. In this construction there are restrictions on which argument can be omitted, as the (d) example shows. The missing argument in the linked core in (4.6a,b) is an actor, in (4.6c,e) an undergoer. In (4.6d), the missing argument has the same semantic role as in the grammatical (e) example; this is crucial evidence that the restriction cannot be stated in semantic role terms. The difference between (4.6d) and (e) concerns the syntactic treatment of the undergoer argument in the linked core. There is thus a restricted neutralization with respect to the omitted NP in (4.6), and it is the same one found in the discussion of verb agreement in (4.1). The privileged syntactic arguments of the two constructions discussed here are the same: they are what would be the core-initial argument in a simple clause, the traditional subject in English. The privileged syntactic arguments of these two constructions are, however, different in an important respect. In (4.1) the privileged syntactic argument is the trigger of verb agreement, while in (4.6) it is the omitted argument in the linked core of a complex sentence. Privileged syntactic arguments thus come in two types; following Heath (1975), the first subtype of privileged syntactic argument will be referred to as a ‘controller’ and the latter as a ‘pivot’. Controllers may trigger verb agreement, as in (4.1), antecede a reflexive, or supply the interpretation for a missing argument in an adjacent unit, as in (4.6). In the matrix cores in (4.6) the NP Chris is a ‘double controller’: it controls both verb agreement on the matrix finite verb, and it controls the interpretation of the missing argument in the linked core. Pivots are canonically (but not exclusively) the missing argument in constructions like (4.6); in other words, the missing argument in the linked core in (4.6a–c,e) is the pivot of the construction. Thus, (4.6c) may be schematically represented as in (4.7); the actual structure of control constructions in terms of the layered structure of the clause will be given in chapter 7. Chrisi wants [—i to drink a beer] controller pivot

(4.7)

In complex constructions, the two commonly co-occur; this is illustrated in (4.8). (4.8) a. a .

Chrisi slapped Patj and then —i/*j ran away. controller pivot Pati was slapped by Chrisj and then —i/*j ran away. controller pivot

95

96


b. b . b .

Chris ran up to the table and — slapped Pat. controller pivot *Chrisi ran up to the table and Pat slapped —i . c ontroller pi vot Chris ran up to the table and — was slapped by Pat. controller pivot

This construction has one privileged syntactic argument in each clause, the controller in the first clause and the pivot (omitted NP) in the second. Each is the traditional subject, as shown by the impossibility of having the undergoer of a transitive verb as the controller in (4.8a) or the pivot in (4.8b ) or of having the actor of a passive verb as controller in (4.8a ). Matrix-coding (or ‘raising’) constructions have a structure which is very similar to that of control constructions; there are two types, and they are illustrated in (4.9). (4.9) a. a . b. b .

It appears that Kim is singing in the shower. Kimi appears [—i to be singing in the shower. pivot Dana believes that Kim is singing in the shower. Dana believes Kimi [—i to be singing in the shower. pivot

Kim is semantically an argument of sing but occurs as the single argument of appear in the matrix core in (a ) and as the ‘direct object’ of believe in (b ); it is the matrix-coded argument of the verb in the linked core (cf. section 7.3.2). The structure in (4.9a ) appears to be the same as in (4.7), but there is a crucial difference: in (4.7), the NP Chris is semantically an argument of want and semantically an argument of drink, whereas in (4.9a ), the NP Kim is not semantically an argument of appear, as (4.9a) shows. Hence there is no controller–pivot relationship in (4.9a ,b ) as in (4.7) and (4.8). The ‘raised’ NP is, rather, analogous to extracted elements in extraction constructions, e.g. WH-question formation in languages like Sama in which there is a restricted neutralization (see (5.3)), which are considered to be pivots. Hence these raising constructions involve a pivot, and what all pivots have in common is that they involve a ‘gap’ in the core containing the predicate of which they are a semantic argument. There are constructions which have just a controller but no pivot. This is exemplified in (4.1) by verb agreement, in which the agreement trigger is a controller, but there is no pivot. Thus the general notion of privileged syntactic argument subsumes pivots and controllers. In discussing grammatical relations, it is useful to characterize restricted neutralizations in the following way. The neutralization of the actor of an intransitive verb and the undergoer of an intransitive verb, as exemplified in (4.1b,c) and (4.6b,c), yields an ‘intransitive subject’ function which may be represented as ‘S’, following Dixon (1972). Acehnese, unlike English, lacks this S function. ‘AT ’ stands for the actor of a transitive verb, and ‘UT ’ stands for the undergoer of


97

a transitive verb. Since a passive verb is a derived intransitive verb in most languages, the single core argument of a passive verb will be termed a ‘derived-S’ [d-S]. The restricted neutralization in (4.1) and (4.6) may be represented as ‘[S, AT , d-S]’, meaning that the single argument of an intransitive verb (regardless of whether it is actor or undergoer), the actor of a transitive verb, and the single argument of a passive verb function alike in a construction. Not all languages show this pattern of neutralization. In Kambera, an Austronesian language of eastern Indonesia (Klamer 1994), the undergoer of a transitive verb cannot function as a pivot in constructions like the one in (4.10). (4.10) a.

Ku-mbuhang pa-ràma. 1sgNOM-want INF-work ‘I want to work.’


b.

Nda ku-mbuhang pa-hidu. NEG 1sgNOM-want INF-be.ill ‘I don’t want to be ill.’


c.

Ku-mbuhang pa-tinung-nya na lau haromu. Actor of transitive V 1sgNOM-want INF-weave-3sgDAT ART sarong tomorrow ‘I want to weave the sarong tomorrow.’

These Kambera control constructions are like the English control constructions in (4.6); there is a missing argument in the linked core. In (4.10a), the omitted argument is the actor of an intransitive verb, while in (b) it is the undergoer of an intransitive verb. Hence Kambera neutralizes the contrast between actor and undergoer with intransitive verbs and, like English, has an S function. With transitive verbs, on the other hand, the situation is rather different. Unlike English, there are no forms like (4.6e) in which the undergoer of a transitive verb functions as a pivot in this construction; Kambera lacks a passive construction. Only the actor of a transitive verb can serve as a pivot in this construction. Hence the neutralization in the Kambera control construction is [S, AT ], not [S, AT , d-S] as in English. Both of these are syntactically accusative patterns. There is another possible neutralization which is found in a number of languages, an ergative one. It is exemplified by the following sentences from Kalkatungu, an Australian Aboriginal language (Blake 1979). These are participial constructions, in which an argument is missing from the participle, which agrees with the controller in case; the issue, as in (4.6) and (4.10), is, which argument can be omitted (serve as pivot) in the linked unit? The linked unit with the pivot is in brackets in both the Kalkatungu example and its English translation. (4.11) a.

b.

Actor of intr. V Tuat.u pa-ji mar.apai-Ø icaji [iŋka-ʎ -iŋka-cin-Ø]. ¯ snake.ERG that-ERG woman-ABS bite go-LNK-go-PART-ABS ‘The snake bit the womani [as —i was walking along].’ [Jar.ikajan-ati-in-tu] caa ŋa-tu laji Ø. ¯ ¯ hungry-VBLZ-PART-ERG here 1sg-ERG kill 3sgABS ‘[—i Being hungry] Ii killed it.’

Undergoer of intr. V

98


c.

d.

Kuntu caa l.uŋa-na Ø [ŋa-tu la-in-ka-Ø]. Undergoer of trans. V ¯ ¯ ¯ NEG here cry-PAST 3sgABS 1sg-ERG hit-PART-SUFF-ABS ‘Hei didn’t cry [when I hit —i ].’ [ar.i-in-Ø katir-Ø]. *Actor of trans. V [active] *Nga-tu n.aa macumpa-Ø ¯ ¯ 1sg-ERG saw kangaroo-ABS eat-PART-ABS grass-ABS ‘I saw the kangarooi [—i eating grass].’

In (4.11a) the missing argument in the participle is the actor of an intransitive verb, while in (b) it is the undergoer of an intransitive predicate; these first two sentences show that Kalkatungu neutralizes actor and undergoer with intransitive verbs and hence has an S function. In (4.11c) the omitted argument in the participle is the undergoer of a transitive verb. The actor of an active voice transitive verb cannot be omitted, as the ungrammaticality of (4.11d) shows. This is analogous to (4.6d), and, like English, there is a way to allow the other macrorole argument of a transitive verb to function as the privileged syntactic argument of the construction. English has a voice construction which allows the undergoer of a lexically transitive verb to serve as the privileged syntactic argument, and Kalkatungu has one which allows the actor of a lexically transitive verb to serve as the privileged syntactic argument; it is termed an ‘antipassive’. The Kalkatungu antipassive construction is exemplified in (4.12); in the active voice in (4.12a), the actor is in the ergative case and the undergoer in the absolutive, while in the antipassive form in (4.12b) the actor is in the absolutive case, functioning as a d-S, and the other argument appears in the dative case. (4.12) a.

b.

Nga-tu la-mi tuar-Ø. ¯ ¯ ¯ 1sg-ERG hit-FUT snake-ABS ‘I’m going to hit the snake.’ Ngai la-ji-mi tuar-ku. ¯ ¯ 1sgABS hit-ANTI-FUT snake-DAT ‘I’m going to hit the snake.’

When the verb in (4.11d) is antipassivized, the actor functions as the single argument of a derived intransitive verb and may be omitted in the construction, as shown in (4.13). (4.13)

katir-ku]. Nga-tu n.aa macumpa-Ø [ar.i-li-in-Ø ¯ ¯ 1sg-ERG saw kangaroo-ABS eat-ANTI-PART-ABS grass-DAT ‘I saw the kangarooi [—i eating grass].’ Actor of transitive V [antipassive]

Hence in the Kalkatungu participial construction the neutralization pattern is [S, UT , d-S]. Note that there is no restricted neutralization with respect to the controller in these constructions: the controller is the UT in (4.11a) and (4.13), the AT in (4.11b), and the S in (4.11c). These neutralization patterns may be summarized as in Table 4.1.6 6

There is one more possible neutralization pattern, [S, UT ]. It is exceedingly rare; see Van Valin and LaPolla (1997), section 6.5, for an explanation of its limited distribution.


Table 4.1 Patterns of restricted neutralizations of semantic roles

Acehnese English Kambera Kalkatungu

Intransitive Vs

Transitive Vs

Grammatical relations

PSA(s)

NO YES YES YES

NO YES NO YES

NO YES YES YES

[A], [U] [S,AT ,d-S] [S,AT ] [S,UT ,d-S]

It might appear that ‘privileged syntactic argument’ is just another term for syntactic subject, but this is not the case. First, privileged syntactic arguments are construction-specific, while grammatical relations like subject are not. It makes no sense to talk about ‘the privileged syntactic argument in English syntax’ in the same way one might talk about ‘the subject in English syntax’, since subject is assumed to be a general relation playing a role in the grammar as a whole; it would, however, be reasonable to discuss ‘privileged syntactic arguments in English syntax’. Conversely, it is nonsensical to talk about ‘the subject of the raising construction in English’, while ‘the privileged syntactic argument of the raising construction in English’ is perfectly natural. Second, the mistaken equation of syntactic subject with privileged syntactic argument is the result of the fact that in languages like English the privileged syntactic argument of almost all of the major constructions appears to be the same.7 There are, however, many languages in which this kind of syntactic consistency is lacking in the most striking way, e.g. Jakaltek (Van Valin 1981) or Sama (Walton 1986); in Jakaltek, for example, there are five different privileged syntactic arguments for the seven major grammatical constructions surveyed in Van Valin (1981). For these languages the assumption that there is a single notion of subject operative in the grammatical system is extremely problematic. Issues like ‘which argument is the syntactic subject in a language like Jakaltek?’ simply cannot arise in this framework. While Jakaltek and Sama point up the difficulties with the assumption that syntactic relations in a language can be described in terms of a single set of generally applicable notions, it is nevertheless the case that most languages are more like English than Jakaltek in having the same privileged syntactic argument for most, if not all, of their major syntactic constructions. Subject, then, is a generalized privileged syntactic argument in languages in which most or all of the major constructions have the same restricted neutralization. It is because of this consistency that languages can be characterized as syntactically accusative or ergative. That is, a language is syntactically accusative if the privileged syntactic argument for the majority of constructions in the grammar treats the actor in clauses with transitive verbs the same as the single argument in clauses with intransitive verbs, i.e. [S, AT ] or [S, AT , d-S]; this is the case in English, for 7

Two constructions which do not have the traditional subject as the privileged syntactic argument are tough-movement, e.g. This book is easy to read, and purpose clauses, e.g. Kim brought the book for the teacher to read.

99

100


example. Since transitive verbs have both an actor and an undergoer, this entails the existence of an accessibility hierarchy to privileged syntactic argument, with actor being the unmarked or default choice and undergoer the marked choice. This markedness is signalled in English and many other syntactically accusative languages by means of the passive construction in which the undergoer of a transitive verb is the privileged syntactic argument. In a syntactically ergative language such as Kalkatungu or Sama (Walton 1986), the markedness hierarchy for accessibility to privileged syntactic argument is reversed: undergoer is the unmarked choice and actor the marked choice. In these languages an antipassive construction exists which permits the actor to function as privileged syntactic argument, e.g. the Kalkatungu antipassive in (4.12b). This is summarized in Table 4.2. Table 4.2 PSA defaults in accusative and ergative systems Syntactic system

Default choice for PSA

Choice for PSA requiring special construction

Accusative Ergative

Actor Undergoer

Undergoer [Passive] Actor [Antipassive]

The markedness of privileged syntactic argument choice is captured in RRG in terms of the privileged syntactic argument selection hierarchy, given in (4.14) and the accessibility to privileged syntactic argument principles in (4.15). (4.14)

(4.15) a. b. c.

d.

Privileged syntactic argument selection hierarchy: arg. of DO > 1st arg. of do > 1st arg. of pred (x, y) > 2nd arg. of pred (x, y) > arg. of pred (x) Accessibility to privileged syntactic argument principles Accusative constructions: highest ranking direct core argument in terms of (4.14) (default) Ergative constructions: lowest ranking direct core argument in terms of (4.14) (default) Restrictions on PSA in terms of macrorole status: 1. Languages in which only macrorole arguments can be PSA: German, Italian, Dyirbal, Jakaltek, Sama, . . . 2. Languages in which non-macrorole direct core arguments can be PSA: Icelandic, Georgian, Japanese, Korean, Kinyarwanda, . . . Restrictions on PSA in terms of coding (Bickel 2003a) 1. Languages with case-sensitive PSAs, e.g. English, German, Nepali, Maithili . . . 2. Languages with case-insensitive PSAs, e.g. Belhare, Tibetan, . . .

(The contrasts in (4.15c,d) will be discussed in sections 4.4 and 4.5.) Even though the hierarchy in (4.14) is similar to the actor–undergoer hierarchy, in that it refers to the same argument positions in LSs, it differs in that it is unilateral, taking agent (argument of DO) as the highest ranking and patient (argument of pred

4.3 Privileged syntactic arguments: types

(x)) as the lowest ranking. If a verb takes both an actor and undergoer (i.e. is M-transitive), the actor will be the highest ranking in terms of (4.14), because the highest ranking argument on the actor–undergoer hierarchy, the actor, will also be the highest ranking argument on (4.14). With an M-intransitive verb, on the other hand, the single macrorole is the highest ranking (for the purposes of (4.15a)) or the lowest ranking (for the purposes of (4.15b)). Hence the single macrorole will be privileged syntactic argument, regardless of which macrorole it is. As Table 4.2 shows, the principles in (4.15a) and (4.15b) may express the default selections in languages with voice oppositions. Note that it makes no sense to talk about an accessibility to privileged syntactic argument hierarchy for Acehnese, because the restrictions refer to actor only, undergoer only, or any core argument.

4.3

Types of privileged syntactic arguments

The data from Acehnese, English, Kambera and Kalkatungu have illustrated several types of pivots and controllers. In the Acehnese control construction in (4.3), the controller in the matrix core is the actor, and the pivot in the linked core is also the actor, while in the possessor raising construction in (4.4), the raised argument is the undergoer. Since these pivots and controllers are either actors or undergoers, they will be termed ‘semantic pivots and controllers’. They are not syntactic in the sense that they do not involve any kind of neutralization, unlike the other pivots and controllers discussed in section 4.2; they do, however, play an important role in the syntax. Pivots and controllers that involve one of the restricted neutralizations in Table 4.1 are syntactic pivots and controllers. Thus the pivots in the English and Kambera control constructions and in the Kalkatungu participial construction are all syntactic pivots. The controller of verb agreement in English in (4.1) is an example of a syntactic controller, as is the controller in the coordinate constructions in (4.8); both exhibit the [S, AT , d-S] pattern typical of English syntax. English does have semantic controllers like Acehnese in some constructions. Consider the control constructions with persuade in (4.16). (4.16) a. b.

Chris persuaded Pati controller Pati was persuaded by Chris controller

[—i to visit Leslie]. pivot [—i to visit Leslie]. pivot

In this control construction, the pivot is the same as in (4.6), but the controller is always the undergoer with persuade: Pat is the undergoer of persuade and is the controller in both sentences, regardless of whether it is syntactically the ‘subject’ or ‘direct object’ (see section 7.3.1). Contrast this with the construction in (4.8), in which the controller is the ‘subject’ regardless of whether it is an actor, as in (4.8a), an undergoer, as in (4.8a ), or an intransitive actor as in (4.8b). Hence

101

102


(4.16) is an example of a semantic controller in English. This contrast may be summarized as in Figure 4.1. Privileged syntactic arguments: pivots and controllers Syntactic Semantic e.g. Kambera [S, AT ] e.g. Acehnese [A], [U] English [S, A T, d-S] English [U] Kalkatungu [S, UT , d-S]

Figure 4.1 Types of privileged syntactic arguments (preliminary)

There is an important difference between the privileged syntactic arguments in English and Kalkatungu, on the one hand, and those in Kambera, on the other. In the Kambera construction the speaker has no choice as to which argument will serve as privileged syntactic argument: if the verb is intransitive, then the single argument will function as privileged syntactic argument, and if the verb is transitive, only the actor can be the privileged syntactic argument. Hence the selection of the privileged syntactic argument is invariable. In English and Kalkatungu, on the other hand, either argument of a transitive verb can function as privileged syntactic argument: there is a default choice, actor for English and undergoer for Kalkatungu, but the existence of a voice opposition in these languages makes it possible for an undergoer in English or an actor in Kalkatungu (of a lexically transitive verb) to serve as privileged syntactic argument in certain constructions. Hence the selection of the privileged syntactic argument is not fixed, as in Kambera, but rather is variable. It is necessary, then, to distinguish variable from invariable privileged syntactic arguments, as in Figure 4.2. The existence of the ‘d-S’ possibility differentiates variable from invariable syntactic pivots and controllers. Privileged syntactic arguments: pivots and controllers Syntactic Variable e.g. English [S, AT, d-S] Kalkatungu [S, UT , d-S]

Semantic e.g. Acehnese [A], [U] English [U] Invariable e.g. Kambera [S, AT ]

Figure 4.2 Types of privileged syntactic arguments (revised)

Given the existence of the variable syntactic controllers and pivots, RRG asks, what factors influence the selection of the argument to be pivot or controller when there is a choice? That is, with a transitive verb, there are in principle two candidates for privileged syntactic argument, the actor and the undergoer,8 and 8

This is something of an idealization, as in many languages direct core arguments which are not undergoers may function as privileged syntactic argument in a passive construction; see sections 4.4–4.5, Van Valin (1991), L. Roberts (1995) and Van Valin and LaPolla (1997) for a detailed examination of this situation in Icelandic. This idealization does not affect the larger point at hand.


given that a choice must be made as to which one will be privileged syntactic argument, what factors, syntactic, semantic and/or pragmatic, govern or at the very least affect this choice? It is essential to recognize that this issue is only relevant to clauses with transitive verbs; there is no question of choice with an intransitive verb, since the single argument may function as privileged syntactic argument. In English, as in many languages, one of the factors influencing privileged syntactic argument selection is discourse pragmatics. It was shown in section 3.2 that in predicate focus constructions, the unmarked focus type, the privileged syntactic argument in English is highly topical, and this suggests that pragmatic factors like topicality could influence the choice of which argument will be selected as the privileged syntactic argument in a clause with a transitive verb. Evidence that this is the case comes from chains of clauses which share a common topical participant; Dixon (1972) labelled these constructions ‘topic chains’. Examples from English are given in (4.17). (4.17) a.

b.

Maryi walked into the department store, —i looked at a couple of dresses, —i bought one, —i went up to the coffeeshop, —i ordered a cup of coffee and —i rested her weary feet. Milti strolled into the casino —i wearing a fake beard, wig and glasses, but —i was immediately recognized by the security people and —i was unceremoniously escorted back out to the street.

Mary is the primary topical participant in (4.17a) and Milt is in (4.17b); they function as privileged syntactic argument in each clause. Mary is an actor in each of the clauses in (4.17a), and accordingly active voice is used in all clauses. In (4.17b), on the other hand, Milt is an actor only in the first two clauses and the undergoer in the last two; passive constructions are used to allow it to function as privileged syntactic argument. It appears, then, that in topic chains like these, the choice of which argument will function as privileged syntactic argument is affected by discourse pragmatic considerations, in particular by a desire to keep the primary topical participant in the unmarked topic position in the clause. All of these clauses are predicate focus constructions, and this is crucial, for it is difficult to have zero anaphora in a clause following a non-predicate focus construction, as the following examples from Lambrecht (1986, 2000) show. (4.18) a. b. c.

There’s Johni and hei ’s reading a book. *There’s Johni and —i is reading a book. Johni is sitting there and —i is reading a book.

(4.19) a. Johni walked into the room and —i spoke to Pat immediately. b. *Into the room walked Johni and —i spoke to Pat immediately. b . Into the room walked Johni and hei spoke to Pat immediately. (4.20) Q: A:

Who married Rosa? a. John i did but hei didn’t really love her. b. ??John i did but —i didn’t really love her.

103

104


(4.21) Q: A:

Did John marry Rosa? Hei did, but hei /—i didn’t really love her.

Example (4.18a) is a presentational construction in which John is focal, and it is not possible to have it as the antecedent for zero anaphora, as (4.18b) shows. The locative inversion construction in (4.19b,b ) is likewise a kind of presentational construction in which the NP John is focal, and here too it cannot be the controller of the pivot in the following clause. In (4.20a), John is a narrow focus (the answer to a WH-question) and likewise cannot be an antecedent for zero anaphora, as (4.20b) shows. Contrast these examples with (4.18c) and (4.21), in which John is clearly established as topical and not focal and can therefore be the antecedent for zero anaphora. Thus in topic chains involving zero anaphora, the selection of the argument to be privileged syntactic argument is strongly influenced by discourse pragmatic factors: it must be the primary topical participant. The following text excerpt illustrates the formation of topic chains in Tepehua, a Totonacan language of Mexico (Watters 1986). (4.22)

‘ˇcun, wa k-tʔ ahun makˇca -n,’ wa nahun ni haciʔ i, yes, FOC 1SUBJ-be.IMPF cook-INF FOC say.IMPF the girl ‘Ka-tawl-ˇciy-ˇca,’ wa jun-kan ni sˇanati1 , mu la -ni-ka- —1 IRR-sit.down-here-already FOC tell-PASS the woman set-DAT-PASS-PRFV ni ʔ iˇs-ta nci, ʔ eˇs wa tawla--ˇca —1 Yu haciʔ i2 tʔ ahun makˇca -n la the 3sg-chair then FOC sit-PRFV-already the girl be.IMPF make-INF very kʔ usi cʔ a lukʔ u, maka -y—2 sˇoqta yu laʔ an laka kuˇsta nice tortilla make-IMPF everything the take.IMPF to cornfield mu la -ta-ˇca —2 mu la -ta —2 mole, tʔ ahni. ʔ eˇs tawa ʔ aqtay--ˇca —2 put.in-PERF-already put.in-PERF mole, turkey and then begin-PRFV-already laka-ni-kan, hun-kan-ˇca —2 cˇ iwinti yu ha ntu laqsawa. gossip-DAT-PASS, tell-PASS-already words that not true ‘Yes, I’m cooking,’ says the girl. ‘Sit down,’ the woman is told, was set a chair and then sat down. The girl was cooking real nice tortillas, [she] makes everything that [she] takes to the cornfield, put in [a basket]; [she] put in mole and turkey, and then [she] began to be gossiped to, is told words that are not true.’

The first sequence of linked clauses has ni sˇanati ‘the woman’ as its primary topical participant. It is the privileged syntactic argument of two passive clauses, junkan ‘be told’ and mulanikan ‘be set [something]’, and an intransitive verb, tawla- ‘sit’. The verb morphology indicates that the participant is the undergoer of ‘tell’ and ‘set’, and the actor of ‘sit’. In the remainder of the passage, ni haciʔi ‘the girl’ is the primary topical participant and is the privileged syntactic argument of all of the remaining clauses. It is the actor in all but the last two clauses, as signalled by the active voice of the verbs, and then in the last two clauses the verbs are passive, indicating a change in semantic function from actor to undergoer. In both English and Tepehua, then, one of the major factors motivating the selection of the argument to function as privileged syntactic argument is discourse pragmatics, in particular whether the argument in question is the primary


topical participant in the context.9 This situation is found in many languages, e.g. Sama (Walton 1986), Tagalog (Foley and Van Valin 1984), Malagasy (Keenan 1976), Dyirbal (Dixon 1972; Van Valin 1981), Jakaltek (Datz 1980), Tzutujil (Butler and Peck 1980), Lango (Noonan and Bavin-Woock 1978). It must be emphasized that pragmatics plays a role in assigning an argument to be the privileged syntactic argument of the relevant construction only with transitive verbs; with intransitive verbs there is no choice, and therefore pragmatics has no role to play. That is, discourse-pragmatics can affect privileged syntactic argument selection only in constructions with variable pivots and controllers. Thus, it is necessary to distinguish pragmatically influenced variable privileged syntactic arguments from those that are not pragmatically influenced; an example of the latter would be the variable syntactic pivot in (4.6), where the selection of the argument in the linked core is driven by the requirement that the pivot be identical to the controller argument in the matrix core, regardless of the context. The final version of the typology of privileged syntactic arguments is given in Figure 4.3. Privileged syntactic arguments: pivots and controllers Syntactic

Semantic e.g. Acehnese [A], [U] English [U] Variable Invariable e.g. Kambera [S, A T] +pragmatic influence −pragmatic influence e.g. English [S, A T, d-S] in (4.17)–(4.21) e.g. English [S, A T, d-S] in (4.6) Tepehua [S, AT , d-S] in (4.22) Kalkatungu [S, UT , d-S] in (4.11), (4.13)

Figure 4.3 Types of privileged syntactic arguments (final)

It should be noted that pragmatically influenced variable privileged syntactic arguments include both controllers and pivots. That is, in topic chains, for example, both the controllers and pivots are pragmatically influenced; indeed, the pivot of the second clause is the controller of the pivot in the subsequent clause, and so on, and it was argued that the motivation for the choice of privileged syntactic argument in each clause is to keep the primary topical participant as privileged syntactic argument in each clause. Hence the topic chains in English in (4.17) and in Tepehua in (4.22) illustrate both pragmatically influenced variable controllers and pivots. It was noted earlier that the traditional notion of ‘subject’ is a generalization across the pivots and controllers of the major syntactic constructions in a language. Given the distinctions in Figure 4.3, it is possible to sharpen this idea a bit: namely, only syntactic pivots and controllers contribute to the generalized privileged syntactic argument notion that is the traditional ‘subject’ in a language. Semantic controllers and pivots do not. This can be seen clearly in the contrast 9

See Branigan and Prat-Sala (2000) and Heydel and Murray (2000) for cross-linguistic psycholinguistic evidence of a discourse motivation for passives in certain contexts.

105

106


between (4.8) and (4.16a); the syntactic controller in the former is the traditional ‘subject’, while the controller in (4.16a) is not. The situation regarding privileged syntactic argument selection in languages with variable privileged syntactic arguments is not, however, the norm universally. In many languages, there is no choice with respect to which argument will be privileged syntactic argument with a transitive verb: it is always the actor. Discourse-pragmatic considerations have no influence on privileged syntactic argument choice. Van Valin and LaPolla (1997), section 6.5, argued that this grammatical relations pattern is in fact the most common pattern cross-linguistically. It is illustrated in (4.23), a switch-reference construction from Amele, a Papuan language (J. Roberts 1988). (4.23)

Ho busale-ce-b dana age qo-ig-a. pig run.out-DS-3sg man 3pl hit-3pl-TPAST ‘The pig ran out and the men killed it.’

In order to form the Amele equivalent of the topic chains in English and Tepehua, the switch-reference construction must be employed to string together clauses joined by switch-reference morphology on the non-final verbs. In this construction markers on dependent medial verbs signal identity or non-identity of particular arguments in the adjacent linked clauses, and these arguments are the privileged syntactic arguments in the construction. Switch-reference systems almost always have a privileged syntactic argument composed of the actor of transitive verbs and the single argument of intransitives, i.e [S, AT ].10 Assume, for the sake of the discussion, that the pig is the primary topical participant in the discourse. In the first clause in (4.23), ho ‘the pig’ is the actor of an intransitive verb, and in the second it is the undergoer of a transitive verb. In sharp contrast to English and Tepehua, in Amele there is no way for ho ‘the pig’ to function as privileged syntactic argument of the final clause in (4.23); only dana age ‘the men’, the actor of qo- ‘hit, kill’, can be privileged syntactic argument. There is no Amele construction analogous to the English topic chain the pig ran out and was killed by the men, in which an undergoer serves as privileged syntactic argument. Thus discourse considerations do not affect the choice of which argument will be privileged syntactic argument with a transitive verb in Amele; only the actor can be the privileged syntactic argument with a transitive verb. Even though the actor will always be the privileged syntactic argument with a transitive verb, there is nevertheless a restricted neutralization of semantic roles for syntactic purposes here, because the construction does not distinguish between actor and undergoer with intransitive verbs. This is what crucially differentiates languages like Amele from languages like Acehnese in which there is no restricted neutralization with intransitive verbs. Other languages which have a system in which discourse-pragmatic factors have no effect on 10

Barai, another Papuan language, and Eastern Pomo, a Hokan language of California, are exceptions to this generalization (for different reasons). See Olson (1978, 1981) and McLendon (1978) for detailed discussion.

4.4 Case marking and agreement

the selection of the argument to be privileged syntactic argument with a transitive verb include Lakhota (Siouan), Choctaw (Muskogean), Cree (Algonquian, North America), Zapotec (Oto-Manguean, Mexico), Enga, Kewa, Yimas, Fore (Papua-New Guinea), Mparntwe Arrernte, Warlpiri (Australia), Chechen, Ingush (Caucasus, Russia), Thai and Vietnamese. It must be emphasized that, within a single language, different constructions may have different privileged syntactic argument selection criteria. That is, in some constructions pragmatic considerations play a role in privileged syntactic argument selection, e.g. determining the pivot argument in the non-initial clauses in topic chains as in (4.17) and (4.22) (where both the controller and the omitted arguments are clause-internal), while in others, e.g. the control construction in (4.6), they are irrelevant and other considerations determine the argument to be privileged syntactic argument. English, for example, has both semantic controllers, as in (4.16) and variable syntactic controllers, as in (4.8) and (4.17). Thus there are two very different situations regarding selection of the argument to function as privileged syntactic argument in syntactic constructions in the world’s languages. In some constructions in some languages, discourse-pragmatic considerations influence this selection, while in the majority of languages there are no constructions in which they play a role. In markedness terms, it could be described as [±pragmatic influence], with [+pragmatic influence] the marked case and [−pragmatic influence] the unmarked case. Since focus (information) structure is a feature of the grammar of all languages, what this contrast represents is the variable grammaticalization of discourse-pragmatic relations in clause-internal morphosyntax, in particular in clause-internal relational structure. As mentioned earlier, whether a privileged syntactic argument is [±pragmatic influence] can only be determined by looking at clauses with transitive verbs, since there is no choice with respect to which argument will be the privileged syntactic argument with an intransitive verb. It is crucial to realize that the issue with respect to whether a privileged syntactic argument is [±pragmatic influence] is not whether the NP functioning as privileged syntactic argument is pragmatically salient or not. Rather, the issue is whether these pragmatic considerations are a factor in the clause-internal syntactic process of selecting the argument to function as privileged syntactic argument, and here Tepehua and Amele differ dramatically: in Tepehua they may affect the selection of the privileged syntactic argument, while in Amele they cannot.

4.4

Case marking and agreement

In syntactic theories in which grammatical relations play a role, case marking and agreement are invariably tied to them, either directly or indirectly. Since notions like subject and direct object have no place in RRG, it is obvious that neither case marking nor agreement can be based on them. Rather, case

107

108


marking rules make crucial reference to macroroles and direct core argument status. The inherent lexical content of NPs (Silverstein 1976, 1981, 1993), which is sometimes misleadingly termed ‘animacy’, also plays an important role in the formulation of case marking rules in many languages. Agreement rules make primary reference to macroroles, although here again inherent lexical content may play a role, e.g. in some languages verbs only agree with animate arguments (Lakhota is such a language). The finite verb agreement rule for Russian, English, Icelandic and German is given in (4.24). (4.24)

Finite verb agreement in Russian, German and Icelandic: the controller of finite verb agreement is the highest ranking macrorole argument (in terms of (4.14)).

This correctly picks out the agreement controller in the English sentences in (4.1), as well as (4.18a) and (4.19b ). The casemarking rules cover regular (non-idiosyncratic11 ) case marking and are given in (4.25) and (4.26); they apply to direct core arguments only. (4.25) a. b. (4.26) a. b.

Case assignment rules for accusative constructions: Assign nominative case to the highest ranking macrorole argument (in terms of (4.14)). Assign accusative case to the other macrorole argument. Case assignment rules for ergative constructions: Assign absolutive case to the lowest ranking macrorole argument (in terms of (4.14)). Assign ergative case to the other macrorole argument.

The application of the rules in (4.25) can be seen in the following Russian examples. (4.27) a.

a . b.

11 12

Uˇcitel’nic-a pro-ˇcita-l-a knig-u. teacher-FsgNOM PRFV-read-PAST-Fsg book-FsgACC ‘The teacher read the book.’ do (uˇcitel’nic-, [read (uˇcitel’nic-, knig-)]) & INGR consumed (knig-)12 Uˇcitel’nic-a govori-l-a. teacher-FsgNOM speak-PAST-Fsg ‘The teacher was speaking.’

For RRG analyses of idiosyncratic or ‘quirky’ case in Icelandic, Latin, Croatian and Hindi, see Van Valin (1991), Michaelis (1993), Dahm-Draksic (1997) and Narasimhan (1998). In logical structures of sentences from languages with morphological case on nouns, the stem of the head noun is represented in the logical structure argument position. Pronouns are represented differently depending on their morphological properties. In languages like Warlpiri, for example, pronouns seem to have an invariable stem and take the same case suffixes as nouns; hence they will be represented by the stem in logical structure, as in (4.28a ). In languages like English and Icelandic, on the other hand, in which pronouns are irregular morphologically, they will be represented in terms of their relevant features, e.g. person, number and gender.


b .

[do (uˇcitel’nic-, [speak (uˇcitel’nic-)])]

c.

ˇ scˇ in-a Zenˇ umer-l-a. woman-FsgNOM die-PAST-Fsg ‘The woman died.’

c .

BECOME dead (ˇzenˇscˇ in-)

The first example contains the M-transitive verb proˇcitat’ ‘read [completely]’; the actor uˇcitel’nic- ‘teacher’ is the highest ranking macrorole and therefore receives nominative case, while the undergoer knig- ‘book’ is the other macrorole and therefore receives accusative case. The last two examples contain M-intransitive verbs, one of which takes an actor (govorit’ ‘speak’ in (4.27b)) and the other an undergoer (umeret’ ‘die’ in (c)). Since the single macrorole, regardless of type, is the highest ranking by virtue of being the only one, it receives nominative case. The application of the ergative case assignment rules in (4.26) is illustrated in the Warlpiri examples in (4.28) from Hale (1973); these rules are the basic ones for an ergative system, and they do not deal with the complexities that arise in split-ergative systems (see Van Valin and LaPolla 1997, section 7.3.1.2). (4.28) a.

Ngaju-Ø ka-rna purla-mi. 1sg-ABS PRES-1sg shout-NONPAST ‘I am shouting.’

a .

[do (ngaju-, [shout (ngaju-)])]

b.

Ngaju-Ø ka-rna mari-jarri-mi. 1sg-ABS PRES-1sg grief-being-NONPAST ‘I feel sorry’, ‘I grieve.’

b .

feel (ngaju-, [grief ])

c.

Ngaju-rlu ka-rna-Ø wawiri-Ø pura-mi. 1sg-ERG PRES-1sg-3sg kangaroo-ABS cook-NONPAST ‘I am cooking the kangaroo.’

c .

[do (ngaju-,)] CAUSE [BECOME cooked (wawiri-)]

The first two examples contain M-intransitive verbs, purla- ‘shout’ (activity) and mari-jarri ‘feel sorry, grieve’ (state); the first takes an actor argument, the second an undergoer. Each is the lowest ranking macrorole in the core and therefore receives absolutive case, following (4.26a). In (c) with an M-transitive verb pura‘cook’, wawiri- ‘kangaroo’ is the undergoer and the lowest ranking macrorole, while ngaju- ‘I’ is the actor and the other macrorole. Hence wawiri- appears in the absolutive case and ngaju- in the ergative case. The rules for assigning dative and instrumental case are the same for both ergative and accusative systems. They are given in (4.29).

109

110


(4.29) a.

Assign instrumental case to non-MR b argument if, given two arguments, a and b, in a logical structure, with (1) both as possible candidates for a particular macrorole and (2) a is equal or higher (to the left of b) on the AUH, b is not selected as that macrorole. Assign dative case to non-macrorole arguments (default).

b.

The idea that dative is the default case for non-macrorole core arguments comes from Silverstein (1981). Instrumental is not a default case and has specific conditions on its application. The assignment of instrumental case in Croatian and Dyirbal is illustrated below. (The Croatian example is from Ranko Matasović, personal communication; the Dyirbal example is from Dixon (1972).) (4.30) a.

b. (4.31) a.

b.

ˇ Zena-Ø je otkljuˇca-l-a vrata kljuˇc-em. Croatian woman-FsgNOM be.3sg unlock-PAST-FEM door.NplACC key-MsgINST ‘The woman unlocked the door with the key.’ [do (ˇzena-, Ø)] CAUSE [[do (kljuˇc-, Ø)] CAUSE [BECOME NOT locked (vrat-)]] Bala yugu-Ø baŋgul yaa-ŋgu baŋgu bari-ŋgu nudi-n. Dyirbal NM.ABS tree-ABS NM.ERG man-ERG NM.INST axe-INST cut-TNS ‘The man cut the tree with an axe.’ [do (yaa-, Ø)] CAUSE [[do (bari-, Ø)] CAUSE [BECOME cut (yugu-)]]

In both of these logical structures, there are two effector arguments, zˇena- ‘woman’ and kljuˇc- ‘key’ in (4.30b) and yaa- ‘man’ and bari- ‘axe’ in (4.31b); both are potential actors. The first effector in each logical structure is chosen as actor, leaving the second effector as a non-macrorole argument. This satisfies the conditions specified in (4.29a), and therefore instrumental case is assigned to the non-actor effectors in both sentences. The application of the dative rule can be exemplified with sentences from Russian and Warlpiri. (4.32) a.

b.

Uˇcitel’nic-a da-l-a knig-u zˇ enˇscˇ in-e. teacher-FsgNOM give-PAST-Fsg book-FsgACC woman-FsgDAT ‘The teacher gave a book to the woman.’ [do (uˇcitel’nic-, Ø)] CAUSE [BECOME have (ˇzenˇscˇ in-, knig-)]

The linking of actor and undergoer is the same as in (4.27a): uˇcitel’nic- ‘teacher’ is the highest ranking argument and therefore the actor, while knig- ‘book’ is the lowest ranking argument and therefore the undergoer. They receive the same cases as in (4.27a). The third core argument, zˇenˇscˇ in- ‘woman’, is a non-macrorole core argument and because the conditions for instrumental case do not apply, it receives dative case. (4.33) a.

b.

Ngaju-rlu ka-rna-ngku karli-Ø yi-nyi nyuntu-ku. 1sg-ERG PRES-1sg-2sg boomerang-ABS give-NONPAST 2sg-DAT ‘I am giving you the boomerang.’ [do (ngaju-, Ø)] CAUSE [BECOME have (nyuntu-, karli-)]


In the Warlpiri example, the ditransitive verb yi- ‘give’ heads the core; ngaju- is the actor and karli- ‘boomerang’ is the undergoer, following the actor–undergoer hierarchy, and nyuntu- ‘you’ is left as a non-macrorole core argument. Accordingly, ngaju- receives ergative case, karli- absolutive case, and nyuntu- dative case. The dative and instrumental rules interact in an interesting way with verbs which permit variable undergoer choice, and such verbs are found in both Croatian and Dyirbal. Examples illustrating this alternation are given in (4.34) from Croatian (Zovko 2000, 2001) and in (4.35) from Dyirbal (Dixon 1972). (4.34) a.

b.

(4.35) a.

b.

darovati ‘give as a gift’ [do (unuk-, Ø)] CAUSE [BECOME have (bak-, cvijet-)] Unuc-i su bak-i darova-l-i cvijeć-e. grandson-MplNOM be.3pl grandmother-FsgDAT give-PAST-pl flower-MplACC ‘The grandsons gave flowers to [their] grandmother.’ Unuc-i su bak-u darova-l-i cvijeć-em. grandson-MplNOM be.3pl grandmother-FsgACC give-PAST-pl flower-MplINST ‘The grandsons gave [their] grandmother flowers.’ wugal ‘give’: [do (d¸ugumbil-, Ø)] CAUSE [BECOME have (yaa-, mira-)] baŋgun d¸ugumbi-u wuga-n bagul yaa-gu. Balam mira-Ø NM.ABS beans-ABS NM.ERG woman-ERG give-TNS NM.DAT man-DAT ‘The woman gave beans to the man.’ wuga-n baŋgun d¸ugumbi-u baŋgum mira-d¸u. Bayi yaa-Ø NM.ABS man-ABS give-TNS NM.ERG woman-ERG NM.INST beans-INST ‘The woman gave the man beans.’

The Croatian verb darovati ‘give as a gift’ allows either the recipient or the theme to function as undergoer. In the default linking, illustrated in (4.34a), cvijet- ‘flowers’, the lowest ranking argument, is selected as undergoer, and bak‘grandmother’, the non-macrorole core argument, receives dative case. In the other linking, shown in (4.34b), the recipient bak- ‘grandmother’ is chosen as undergoer, leaving cvijet- ‘flowers’ as a non-macrorole argument. This linking meets the conditions of the instrumental case rule: both cvijet- ‘flowers’ and bak- ‘grandmother’ are potential undergoers, and bak- ‘grandmother’ is to the left of cvijet- ‘flowers’ in the logical structure. Consequently, cvijet- ‘flowers’ is assigned instrumental case. The same contrast is found in the Dyirbal examples in (4.35). In the first sentence, the lowest ranking argument, mira- ‘beans’, is selected as undergoer, leaving yaa- ‘man’ as a non-macrorole argument; it receives dative case. In (4.35b), however, the recipient yaa- ‘man’ is chosen as the undergoer, leaving mira- ‘beans’ a non-macrorole argument. Here too the instrumental rule can apply, because yaa- ‘man’ is higher on the actor– undergoer hierarchy than mira- ‘beans’; accordingly, mira- ‘beans’ receives instrumental case. In these pairs of sentences, dative case on the non-macrorole argument correlates with the unmarked choice for undergoer in terms of the actor– undergoer hierarchy, while instrumental case on it correlates with a marked choice

111

112


for undergoer, i.e. the undergoer is not the lowest ranking argument in the logical structure. The main point for this discussion is that the rules in (4.25) and (4.26) make no reference to grammatical relations of any kind at all; indeed, they do not even refer directly to the notion of privileged syntactic argument. This is quite important, because in Icelandic, as in many languages, the principles determining the privileged syntactic argument are independent of the case assignment principles. This can be seen most clearly in examples like the following. ´ Mér hef-ur þo´ tt Olaf-ur leið inleg-ur. 1sgDAT have-3sg think.PSTP Olaf-MsgNOM boring-MsgNOM ‘I considered Olaf boring.’ ´ þo´ tt Olaf-ur leið inleg-ur. b. Mér virð ist hafa 1sgDAT seem have.INF think.PSTP Olaf-MsgNOM boring-MsgNOM ‘I seem to have considered Olaf boring.’ ´ virð ist þeim hafa þo´ tt leið inleg-ur. b . *Olaf-ur Olaf-MsgNOM seem 3plDAT have.INF think.PSTP boring-MsgNOM *‘Olaf seems to have been considered boring by them.’

(4.36) a.

The logical structure for (4.36) is consider (1sg, [be (Olaf, [boring ])]) [MR 1]; because this verb is M-intransitive, the single macrorole is undergoer, following the principles in (2.41b), and the other argument is a non-macrorole core argument. ´ Hence Olafis the undergoer, and the first singular pronoun is a non-macrorole core argument. With respect to the privileged syntactic argument selection hierarchy in (4.14), ‘1sg’ is the highest ranking argument and therefore would be the privileged syntactic argument, and this is confirmed by the grammaticality of (4.36b) and the ungrammaticality of (4.36b ). The case assignment rules, on the ´ other hand, refer to highest ranking macrorole, and in this sentence that is Olaf-, the undergoer, which appears in the nominative case. In sentences like (4.36), the privileged syntactic argument selection hierarchy and the nominative case rule pick out different NPs. In a language like German in which only macroroles can serve as privileged syntactic argument, the nominative NP will always be the privileged syntactic argument, unless it is a dummy es ‘it’, as in, e.g., Es regnet ‘it is raining’. In (4.15d) a restriction in terms of the case-sensitivity of privileged syntactic arguments was introduced; this contrast was proposed in Bickel (2003a). It concerns whether privileged syntactic argumenthood is restricted in terms of the morphological coding of the highest ranking argument in the core. Consider the following contrast in agreement between Nepali, an Indo-European language, and Belhare, a Tibeto-Burman language. (4.37) a.

Ma bhut sa˙nga d.ar¯a-¯e. 1sgNOM ghost with fear-1sgPAST ‘I was afraid of the ghost.’

Nepali


b.

(4.38) a.

b.

Ma-laı bhut sa˙nga d.ar l¯ag-yo/*l¯ag-¯e. 1sg-DAT ghost with fear feel-3MsgPAST/feel-1sgPAST ‘I was afraid of the ghost.’ i? Han-na tombhira kiiʔ -t-u-ga 2sg-ERG lynx.NOMsg fear-NPAST-3sgO-2sgA Q ‘Are you afraid of the lynx?’

Belhare

Ngka hale hani-ni¯ua ka-tiu-s-ik-kha. 1sgNOM before 2plGEN-mind.NOM 1sgO-spend-TR-2plA-PERF ‘Before, you liked me.’

In all four of these sentences, the experiencer is the highest ranking argument, in terms of the privileged syntactic argument selection hierarchy in (4.14), and therefore it should be the privileged syntactic argument in each sentence. This is the case in (4.37a), in which the experiencer appears in the nominative case and triggers first-person-singular agreement on the verb. In (4.37b), on the other hand, the experiencer does not trigger agreement; it appears in the dative case, and the verb bears impersonal inflection. The only difference between the two sentences is the case assigned to the experiencer: nominative in (4.37a) but dative in (b). The dative case in (b) blocks the experiencer from being a controller of verb agreement, hence from being a privileged syntactic argument. The situation in Belhare is strikingly different: the experiencer triggers transitive subject agreement on the verb regardless of its morphosyntactic coding. In (4.38a) it is a direct core argument in the ergative case, while in (b) it is coded as the possessor of a body part term, in this case the noun meaning ‘mind’. Nevertheless, it still triggers second-person-plural transitive subject agreement on the verb, showing that it functions as a privileged syntactic argument. Thus, it is necessary to have case sensitivity as a factor in privileged syntactic argument selection.13 In English, case marking rules like those in (4.25) apply only to pronouns, and the equivalent for lexical NPs is prepositional case marking. As noted in section 2.4.2, the adpositions that mark oblique core arguments are not listed in the lexical entry of the verb but rather are predicted by general principles. This approach to preposition assignment was first presented in Foley and Van Valin (1984) and further developed in Jolly (1991, 1993) and Van Valin and LaPolla (1997). The basic rules governing the assignment of to, from and with are given in (4.39).14 (4.39) a. b.

13 14

Preposition assignment rules for English Assign to to non-MR y argument in logical structure segment: . . . BECOME/INGR pred (y, z) Assign from to non-MR y argument in logical structure segment: . . . BECOME/INGR NOT pred (y, z)

See Bickel (2003a) for the wider syntactic ramifications of this distinction. These rules do not cover all uses of from or with. See Jolly (1991, 1993) for discussion of other uses of from. Van Valin and LaPolla (1997), section 7.3.2.1, discusses a wide range of uses of with.

113

114


c.

Assign with to non-MR b argument if, given two arguments, a and b, in a logical structure, with (1) both as possible candidates for a particular macrorole and (2) a is equal or higher (to the left of b) on the AUH, b is not selected as that macrorole.

The rule for assigning with is basically the same as the rule for assigning instrumental case in (4.29a). The application of these rules for to and with can be illustrated with the sentences in (4.40). They are all concerned with the undergoer end of the actor–undergoer hierarchy and the selection of the undergoer macrorole. The alternation with present in (4.40b,b ) is known as the ‘transfer alternation’, while the one in (4.40c,c ) with load is known as the ‘locative alternation’. (4.40) a. a . b. b . b . c. c . c .

Sally gave the flowers [z] to Kim [y]. Sally gave Kim [y] the flowers [z]. Sally presented the flowers [z] to Kim [y]. Sally presented Kim [y] with the flowers [z]. [do (Sally, Ø)] CAUSE [BECOME have (Kim, flowers)] Max loaded the olives [z] into his minivan [y]. Max loaded his minivan [y] with the olives [z]. [do (Max, Ø)] CAUSE [BECOME be-in (minivan, olives)]

In the LS in (4.40b ), Kim corresponds to the y argument and flowers to the z argument in (4.39a). In (4.40a,b) the flowers is selected as undergoer, leaving Kim as a non-macrorole argument, and by (4.39a) it would be assigned to. The sentence with load in (4.40c) also meets the conditions for the assignment of to to the y argument, but because the predicate in the LS is a prepositional predicate, be-in , to is in effect added to the prepositional predicate to yield into.15 In the (a ), (b ) and (c ) sentences in (4.40), however, the higher ranking argument on the actor–undergoer hierarchy, the y argument, is selected as undergoer instead of the lowest ranking argument, z. In this situation the with rule in (4.39c) should apply, assigning with to the z argument (= b in (4.39c)), because it has not been selected as a macrorole. The primary difference between dative shift verbs like give and transfer verbs like present in English is that the with rule does not apply to the z argument of dative shift verbs. Transfer verbs in English are analogous to the Croatian and Dyirbal verbs discussed above, in that they assign with (the English analogue of instrumental case) to the lowest ranking argument when it is not selected as undergoer. This same rule applies at the other end of the actor–undergoer hierarchy, as illustrated in (2.38), repeated as (4.41).

15

The examples in (4.40c,c ) exhibit the much discussed variation in the affectedness of the olives and the minivan, i.e., in (c) all of the olives are loaded into the minivan, with no implication that the minivan is fully loaded, whereas in (c ) the minivan is fully loaded, with no implication that all of the olives have been loaded. This difference in interpretation is attributed to the different undergoer selections, the undergoer being the most affected participant in the state of affairs. See Van Valin and LaPolla (1997:145).

4.5 Other syntactic functions

(4.41) a. b.

Leslie shattered the window with a rock. [do (Leslie, Ø)] CAUSE [[do (rock, Ø)] CAUSE [INGR shattered (window)]]

In this logical structure Leslie and rock are both effectors and therefore potential actors; indeed, if Leslie were omitted, a rock would be the actor, as in A rock shattered the window. However, in (4.41) Leslie is the effector of the superordinate CAUSE and therefore outranks a rock to be actor. In terms of (4.39c), Leslie is a and a rock is b, and since a rock is not selected as actor, it is marked by with. This is exactly analogous to the assignment of instrumental case in the Croatian and Dyirbal examples in (4.30) and (4.31). Thus the same basic principle governs the assignment of with in both (4.40) and (4.41). The application of the from rule in (4.39b) can be seen in the examples in (4.42). (4.42) a. b. c.

The pool cleaner drained the water [z] from the pool [y]. The pool cleaner drained the pool [y] of its water [z]. [do (pool cleaner, Ø)] CAUSE [BECOME NOT be-in (pool, water)]

In (4.42a) the z argument, the water, is selected as undergoer, and the conditions for the from rule in (4.39b) are met, resulting in the y argument, the pool, being marked by from. In (4.42b), on the other hand, the higher ranking y argument is selected as undergoer, resulting in a marked undergoer selection and the application of the with rule in (4.39c). However, with removal verbs like drain, with is replaced by of.

4.5

Other syntactic functions

It was stated in section 4.0 that RRG does not posit any clause-internal syntactic relations beyond privileged syntactic argument, and therefore there are no relations akin to the traditional notions of direct or indirect object in RRG. ‘Direct object’ is by far the more important of the two relations, and the motivation for postulating such a relation in other theories has come primarily from two sets of phenomena: (1) voice constructions in which an underlying direct object appears as surface subject, and (2) constructions in which new direct objects are added to a clause (‘applicative constructions’). In this section it will be shown that the notions of undergoer and core argument account for these phenomena, thereby rendering ‘direct object’ superfluous. In syntactic theories employing multiple derivationally related syntactic levels of representation, the notion of direct object plays a crucial role in the statement of voice oppositions. In RRG, by contrast, passive is formulated in terms of the hierarchy of privileged syntactic argument selection in (4.15a): passive always involves a marked privileged syntactic argument choice, with the undergoer appearing as privileged syntactic argument in the default situation;

115

116


cf. Table 4.2. No notion of syntactic direct object is involved. There are usually (but not always) two facets of a passive construction, the occurrence of a marked privileged syntactic argument choice, and the omission of the actor or its appearance as an oblique element in the periphery. The former is referred to as ‘PSA modulation’, the latter as ‘argument modulation’, in Van Valin and LaPolla (1997). Antipassive constructions canonically involve these same two phenomena: a marked privileged syntactic argument selection and the omission or non-canonical coding of the undergoer. The universal formulation of the basic voice oppositions is presented in (4.43). (4.43) a. b.

General characterization of basic voice constructions PSA modulation voice: permits an argument other than the default argument in terms of (4.14) to function as the privileged syntactic argument. Argument modulation voice: gives non-canonical realization to a macrorole argument.

An important motivation for factoring voice constructions into these two parts is that they occur independently of each other in some languages. Impersonal passives with intransitive verbs of the kind found in German, Latin, Turkish and many other languages involve only (4.43b), since the resulting constructions have no privileged syntactic argument. There are also languages in which this kind of impersonal passivization applies with transitive verbs, suppressing the actor but leaving the undergoer as a non-privileged syntactic argument in the accusative case, e.g. Ute (Uto-Aztecan; Givón 1981). Here again passive involves only (4.43b) and not (4.43a). These two possibilities are illustrated in (4.44). (4.44) a.

b.

Hier wird nicht gelacht. here become.3sgPRES not laughed ‘No laughing here.’ (lit. ‘here is not laughed.’) Sivá¸ ¸utu-ci ‘u-wáy paˆxa´ -ta-ˆxa. goat-ACC the-ACC kill-PASS-TNS ‘The goat was killed’, or ‘Someone killed the goat.’

German

Ute

On the other hand, in Lango (Nilo-Saharan; Noonan and Bavin-Woock 1978; Noonan 1992) the undergoer may occur as privileged syntactic argument with the actor remaining a direct core argument; this Lango construction involves privileged syntactic argument modulation only, and there is no special morphology on the verb signalling the construction, unlike in the impersonal passives mentioned above. L. Roberts (1995) argues that passive morphology on the verb is associated with argument modulation only, and the contrast between, e.g., Ute and Lango supports her claim. A similar situation is found in many antipassive constructions, in which the actor functions as the privileged syntactic argument of the construction, but the undergoer remains a direct core argument, e.g. Jakaltek (Craig 1977), Sama (Walton 1986). The Lango and Sama constructions are exemplified in (4.45) and (4.46).


(4.45) a.

b.

(4.46) a.

b.

Dákó o` -jwát-ò lócà. woman 3sgA-hit-3sgU man ‘The woman hit the man.’

Lango

Lócà dákó o` -jwát-ò. man woman 3sgA-hit-3sgU ‘The man was hit by the woman.’ B’lli d’nda daing. buy woman fish ‘The woman bought the fish.’

Sama

N-b’lli d’nda daing. ANTI-buy woman fish ‘The woman bought fish.’

The Lango construction in (4.45b) looks like a simple fronting of the undergoer, but it is in fact a voice construction; this is shown by the fact that the fronted NP takes over a number of syntactic control properties from the initial NP in (4.45a). The actor, dákó ‘woman’, remains a direct core argument and continues to control verb agreement and could control a reflexive anaphor. There seems to be no morphological difference between the two Sama sentences except for the nasal prefix on the verb which signals antipassivization. However, syntactically, there is an important difference: for extraction constructions, daing ‘fish’ would be the privileged syntactic argument in (4.46a), but d’nda ‘woman’ would have that function in (b) (see (5.3), (5.4)). However, in (b) daing ‘fish’ remains a direct core argument.16 It was mentioned in (4.15c) that there are many languages in which only undergoers may appear as privileged syntactic argument in a passive construction; German, Italian and Indonesian are three examples. Icelandic presents a very interesting situation in which privileged syntactic argument status in passives is not restricted to the undergoer argument of a transitive verb. The verb in (4.47) is a regular case-marking transitive verb; in (4.47a) the nominative argument is the privileged syntactic argument and is an actor, and the accusative argument is 16

There is an important asymmetry between the argument modulation features of passives and antipassives. It is relatively unusual for passivization not to involve the actor being treated as a peripheral adjunct or omitted altogether, but it appears to be equally unusual for the undergoer in an antipassive to be treated as a peripheral adjunct. There are many examples of antipassives in which the undergoer seems to be completely unaffected, e.g. the Sama example above. In these languages the antipassive has only the PSA modulation component. In other cases, such as the Kalkatungu antipassive in (4.12b), the undergoer of the active voice seems to retain its core status but be stripped of its macrorole status, appearing in the dative case. The explanation for this contrast lies in the different syntactic potential of actor vs undergoer. Actor arguments are powerful syntactically and typically possess many controller and pivot properties, while undergoers typically do not. Hence leaving an undergoer as a direct core argument, regardless of whether it is a macrorole or not, is unlikely to lead to confusions regarding the agreement controller, reflexive antecedent or pivot in certain constructions. On the other hand, having the actor remain a direct core argument leads to potential ambiguity with respect to reflexive control, control of missing arguments in complex constructions, etc. Treating an actor as a peripheral adjunct indicates clearly that it has lost the controller and pivot properties which accrue to the undergoer in the passive construction.

117

118


an undergoer, while in (4.47b) the undergoer is privileged syntactic argument in a passive construction. (4.47) a.

b.

Lögregl-a-n tók Sigg-u fast-a. police-FsgNOM-DEF take.PAST Sigga-FsgACC fast-FsgACC ‘The police arrested Sigga.’ Sigg-a va-r tek-in föst Sigga-FsgNOM be.PAST-3sg take.PSTP-FsgNOM fast.FsgNOM af lögregl-un-ni. by police-DEF-FsgDAT ‘Sigga was arrested by the police.’

Many apparently transitive verbs in Icelandic have ‘direct objects’ which are in the dative or genitive case, not the accusative, and as argued in Van Valin (1991), the undergoer of a transitive verb in Icelandic always receives accusative case in active voice (simple) sentences. From this it follows that none of the postverbal dative NPs in (4.48) are undergoers; hence these verbs are M-intransitive, despite having multiple arguments.17 (4.48) a.

b.

´ Eg skila-ð -i henni pening-un-um. 1sgNOM return-PAST-1sg 3FsgDAT money-DEF-DAT ‘I returned her the money.’ ´ honum. Eg hjálpa-ð -i 1sgNOM help-PAST-1sg 3MsgDAT ‘I helped him.’

The dative rule in (4.29b) applies not only to ditransitive verbs like dat’ ‘give’ in Russian, geben ‘give’ in German, and gefa ‘give’ in Icelandic but also to multiple argument verbs which are M-intransitive. The Icelandic verbs skila ‘give back, return’ and hjálpa ‘help’ in (4.48) are good examples of multiple argument, M-intransitive verbs. Their single macrorole is an actor, which receives nominative case, but their remaining arguments are non-macrorole core arguments, which receive dative case. Traditionally such verbs are analysed as governing irregular case marking on some of their arguments, but in terms of the RRG analysis their only irregularity is in terms of transitivity: the macrorole assignment principles in (2.41a) predict that they should be M-transitive, but they are in fact M-intransitive. This would be indicated in their lexical entry, as discussed in section 2.4.2. But given the fact that they are M-intransitive, the case marking of their direct core arguments is completely regular in terms of (4.25) and (4.29). A dative NP in Icelandic sentences like (4.48) may function as a true privileged syntactic argument in a passive construction, as can be seen in (4.49); in this way

17

See Van Valin and LaPolla (1997:352–61) for detailed arguments in support of the conclusion that dative objects in Icelandic and German are non-macrorole core arguments.


Icelandic differs from German, in which ‘dative subjects’ in passives do not take on the behavioural properties of nominative ‘subjects’.18 (4.49) a.

Henni va-r skil-að pening-un-um. 3FsgDAT be.PAST-IMPER return-PSTP money-DEF-DAT ‘She was returned the money.’

va-r skil-að henni. a . *Pening-un-um money-DEF-DAT be.PAST-IMPER return-PSTP 3FsgDAT ‘The money was returned to her.’ b.

Honum va-r hjálp-að (af mér). 3MsgDAT be.PAST-IMPER help-PSTP (by 1sgDAT) ‘He was helped (by me).’

Evidence that they are true privileged syntactic arguments comes from matrixcoding (raising) constructions. The relevant examples are given in (4.50). (4.50) a.

Lögregl-a-n virð ist hafa tek-ið Sigg-u police-FsgNOM-DEF seem.IMPER have.INF take-PSTP Sigga-FsgACC fast-a. fast-FsgACC ‘The police seem to have arrested Sigga.’

b.

hafa ver-ið tek-in Sigg-a virð ist Sigga-FsgNOM seem.IMPER have.INF be-PSTP take.PSTP-FsgNOM föst af lögregl-un-ni. fast.FsgNOM by police-DEF-FsgDAT ‘Sigga seems to have been arrested by the police.’

c.

hafa ver-ið hjálp-að (af mér). Honum virð ist 3MsgDAT seem.IMPER have.INF be-PSTP help-PSTP (by 1sgDAT) ‘He seems to have been helped (by me).’

In (4.50a,b) the privileged syntactic argument of the linked core appears as the privileged syntactic argument of the matrix core, and the verb in the linked core is the regular case-marking verb illustrated in (4.47). The important example is (4.50c), in which the dative NP (which is the d-S of the passive construction) occurs as the privileged syntactic argument of the matrix core, just like the nominative NPs in (4.50a,b). This is evidence that the dative NP honum ‘him’ in (4.49b) functions as a true privileged syntactic argument with respect to constructions like this one. Since these verbs do not have undergoers, the correct statement of the linking in the passive construction in Icelandic cannot be ‘U = PSA’; rather, it must be ‘∼A = PSA’.19 The characterization of the non-actor argument is an important question, as the ungrammaticality of (4.49a ) shows; with a verb like skila ‘return, 18

19

All Icelandic examples are from Zaenen, Maling and Thráinsson (1985), Thráinsson (1979) and Andrews (1982); see these references for detailed discussion and exemplification of the privileged syntactic argument properties of the non-nominative ‘subjects’. See L. Roberts (1995) for an analysis of passivization in Icelandic and Old English within RRG.

119

120


give back’ which takes two non-actor arguments, only one of them can be privileged syntactic argument in a passive. This verb has roughly the same LS as English give, i.e. [do (x, Ø)] CAUSE [BECOME have (y, z)], and accordingly in (4.48a) e´ g ‘I’ is the x argument, henni ‘her’ is the y argument, and peningunum ‘the money’ is the z argument. In terms of the privileged syntactic argument selection hierarchy in (4.14), henni outranks peningunum, and therefore when there are two non-macrorole arguments, the highest ranking direct core argument will be privileged syntactic argument. That direct core argument and not just core argument is the crucial notion can be seen in (4.51), in which skila appears with three core arguments but only two direct core arguments. (4.51) a.

b. b .

´ Eg skila-ð -i pening-un-um til hennar. 1sgNOM return-PAST-1sg money-DEF-DAT to 3FsgGEN ‘I returned the money to her.’ skil-að til hennar. Pening-un-um hefur ver-ið money-DEF-DAT have.IMPER be-PSTP return-PSTP to 3FsgGEN ‘The money was returned to her.’ *Hennar hefur ver-ið skil-að pening-un-um (til). 3FsgGEN have.IMPER be-PSTP return-PSTP money-DEF-DAT to ‘She was returned the money.’

In (4.51a) the recipient (y) argument is not a direct core argument; it is marked by the preposition til ‘to’ and receives genitive case from it. In (4.51a), peningunum ‘the money’, the z argument, is the highest ranking non-actor direct core argument, while til hennar ‘to her’ is the highest ranking core argument, direct or oblique, but only peningunum can be privileged syntactic argument in a passive, as in (4.51b). Hence only direct core arguments can be privileged syntactic argument in a passive construction. Icelandic thus differs from German in permitting non-macrorole arguments to function as the privileged syntactic argument in certain constructions. This contrast is stated in (4.15c), and it represents an important difference across languages. Thus for languages like German the privileged syntactic argument selection hierarchy is effectively Actor > Undergoer, while for languages like Icelandic it is the hierarchy in (4.14) referring to direct core arguments, which includes macrorole and non-macrorole arguments. With respect to finite verb agreement in both languages, the agreement rule in (4.24) guarantees that the finite verb in German, for example, will agree with the privileged syntactic argument if there is one, since only macrorole arguments can function as privileged syntactic argument in the language. In an Icelandic sentence like (4.36a) the finite verb will agree with the nominative undergoer but not with the dative privileged syntactic argument. The nominative NP and the controller of finite verb agreement will coincide in Icelandic, even though the privileged syntactic argument for syntactic processes like matrix-coding and reflexive controller may be a different NP, as illustrated for matrix-coding in (4.36b,b ). Impersonal agreement of the kind found in (4.49b), for example, is a kind of default agreement which occurs when the rule in (4.24) cannot apply.


The other phenomenon in which ‘direct object’ appears to play a crucial role is constructions in which new objects are derived, i.e. dative shift and applicative constructions. They are traditionally described as involving the promotion of a non-object to direct object status. In RRG terms they involve a marked undergoer assignment in terms of the actor–undergoer hierarchy in Figure 2.4. The first type of alternation, which includes dative shift, the transfer alternation and the locative alternation, was illustrated in the Croatian and Dyirbal examples in (4.34) and (4.35) and in the English examples in (4.40) and (4.42). The second type of marked undergoer construction involves the occurrence of a non-argument of the verb as undergoer, the most frequent one being a beneficiary. This is exemplified in (4.52) for English and in (4.53) for Sama (Walton 1986). (4.52) a. b.

Sandy baked a pie for Robin. Sandy baked Robin a pie.

(4.53) a.

taumpa ma si Andi. B lli ku buy 1sgERG shoes for PM Andy ‘I bought the shoes for Andy.’

b.

si Andi taumpa . B lli-an ku buy-BEN 1sgERG PM Andy shoes ‘I bought Andy some shoes.’

The logical structure for English bake or Sama b lli ‘buy’ does not contain a beneficiary argument, and in the (a) sentences this argument appears as an oblique phrase. In the (b) sentences, however, it appears as undergoer, and in Sama this requires the addition of the -an suffix to the verb. Since Sama is syntactically ergative, the derived undergoer is also the privileged syntactic argument (see Walton 1986 for detailed discussion). This supports the point made above that undergoer, and not direct object, is the relevant notion for the description of these alternations. The occurrence of a beneficiary as undergoer is very common, and in some languages, e.g. many Bantu languages, there is no way to express a beneficiary except as undergoer in the ‘applied verb’ (applicative) construction. Examples from Chichewa ˆ (Baker 1988) and Kinyarwanda (Kimenyi 1980) are given in (4.54) and (4.55).20 (In the glosses, the numbers refer to noun classes.) (4.54) a.

b.

20

Mavuto a-na-umb-a mtsuko. Mavuto 1-PAST-mold-MOOD waterpot ‘Mavuto molded the waterpot.’

Chichewa ˆ

Mavuto a-na-umb-ir-a mfumu mtsuko. Mavuto 1-PAST-mold-APPL-MOOD chief waterpot ‘Mavuto molded the waterpot for the chief.’

The facts of Bantu applicative constructions are very complex, and a complete treatment of them is beyond the scope of this discussion. That the ‘applied object’ is undergoer is shown by its taking on all (Chichewa) ˆ or most (Kinyarwanda) of the properties of undergoers of underived transitive verbs. See L. Roberts (1995) for discussion of applicatives from an RRG perspective.

121

122


(4.55) a.

Umukoôbwa a-ra-som-a igitabo. girl 1-PRES-read-MOOD book ‘The girl is reading a book.’

b.

Kinyarwanda

Umukoôbwa a-ra-som-er-a umuhuûngu igitabo. girl 1-PRES-read-APPL-MOOD boy book ‘The girl is reading the boy a book.’

One of the intriguing features of applicative constructions with beneficiaries is that in some Bantu languages the derived benefactive undergoer as well as the other direct core arguments of the verb can function as privileged syntactic argument in a passive construction. This is illustrated in (4.56) from Kinyarwanda. (4.56) a.

b.

c.

d.

Umugóre a-rá-hé-er-a umugabo ´ımbwa ib´ıryo. woman 1-PRES-give-APPL-MOOD man dog food ‘The woman is giving food to the dog for the man.’ Ib´ıryo bi-rá-hé-er-w-a umugabo ´ımbwa n’ûmugóre. food 8-PRES-give-APPL-PASS-MOOD man dog by-woman ‘The food is being given to the dog for the man by the woman.’ ´ Imbwa i-rá-hé-er-w-a umugabo ib´ıryo n’ûmugóre. dog 4-PRES-give-APPL-PASS-MOOD man food by-woman ‘The dog is being given food for the man by the woman.’ Umugabo a-rá-hé-er-w-a ´ımbwa ib´ıryo n’ûmugóre. man 1-PRES-give-APPL-PASS-MOOD dog food by-woman ‘The man benefits from the woman giving food to the dog.’ (lit. ‘The man is being given food to the dog by the woman.’)

These examples show that ‘∼A’ is considerably more than just ‘U’ in the statement of passive in Kinyarwanda; as in Icelandic, direct core arguments other than the undergoer are possible as privileged syntactic argument in a passive construction. Bantu languages also present a kind of derived ‘object’ construction in which a non-argument becomes a direct core argument but not undergoer. The examples in (4.57) are again from Kinyarwanda; in (4.57b) a non-argument locative and a possessor are coded as direct core arguments. (4.57) a.

b.

´ Umw´ aana y-a-andits-e iz´ıná rye mu igitabo cyuˆ mugabo. child 1-PAST-write-MOOD name of.him in book of man ‘The child wrote his name in the man’s book.’ ´ Umw´ aana y-a-andik-i-yé-mo umugabo igitabo iz´ıná rye. child 1-PAST-write-APPL-MOOD-LOC man book name of.him ‘The child wrote in the man’s book his name.’

There are three direct core arguments after the verb in (4.57b), but, unlike (4.56a), only umugabo ‘man’, the undergoer (the applied ‘object’), has the syntactic properties of the undergoer of a non-derived transitive verb, e.g. the ability to function as privileged syntactic argument in a passive construction. Igitabo ‘book’ is coded


as a peripheral oblique in (4.57a) but as a direct core argument in (4.57b); it is not, however, the undergoer. Applicative constructions (and their cousins, the dative shift, transfer and locative alternations) pose a variety of interesting problems for linguistic theory, but they do not motivate the postulation of a direct object grammatical relation. The notions of undergoer and direct core argument provide the basis for the description and ultimately for the explanation of these phenomena. A further problem for traditional notions of grammatical relations are what Dryer (1986) called ‘primary object languages’; in these languages the only pattern that occurs with three-argument verbs corresponds to (4.40a ), not (4.40a). Lakhota provides a simple example of this. (4.58) a.

Mathó ki hená sˇu´¸ ka eyá wiˇchá-kte-pi. bear the those dog some 3plU-kill-3plA ‘Those bears killed some dogs.’

b.

W´¸ı ya¸ ki mathó wa¸ hokˇs´ıla ki hená wiˇchá-Ø-kipazo/*Ø-Ø-kipázo. woman the bear a boy the those 3plU-3sgA-show/3sgU-3sgA-show ‘The woman showed those boys a bear.’

c.

kipázo ‘show’ [do (w´¸ı ya¸ , Ø)] CAUSE [BECOME see (hokˇs´ıla [y], mathó [z])]

In (4.58a) with a transitive verb, the plural animate undergoer is coded by the prefix wiˇcha- on the verb. With a three-argument verb like kipázo ‘show’, which of the non-actor arguments is coded as the undergoer? As (4.58b) clearly attests, it is not the lowest ranking argument in the LS that functions as undergoer, as would be the case in a construction like (4.40a), but rather the higher ranking y argument. Moreover, there is no construction in Lakhota with a three-argument verb in which the z argument would be the undergoer. Because this pattern is distinct from the traditional ‘direct object’ – ‘indirect object’ pattern of (4.40a), Dryer labels it the ‘primary object’ – ‘secondary object’ pattern, with hokˇs´ıla ki hená ‘those boys’ as the primary object and mathó wa¸ ‘a bear’ as the secondary object in (4.58b). In the English example in (4.40a ), Kim would be the primary object and the flowers the secondary object. In RRG terms, primary object languages allow only the marked selection for undergoer, in terms of the actor–undergoer hierarchy. How can this be explained? There would appear to be two approaches within RRG. The first, put forth in Van Valin and LaPolla (1997), section 7.4.1, maintains that undergoer selection based on the actor–undergoer hierarchy is affected by a general principle to the effect that if there is an animate (human) argument as a potential undergoer, then it must be undergoer. In the usual situation involving transfer verbs, in which the recipient is normally human or animate and the theme inanimate, this correctly predicts that the animate recipient would be selected as undergoer, as in, e.g., (4.40a ). However, examples like (4.58b) present a problem for this analysis. While it is certainly the case that the undergoer in (4.58b) is animate, the non-undergoer theme is also

123

124


animate, and therefore this principle does not really explain examples like this one, even though they do not contradict it. The alternative analysis claims that undergoer selection is based on the principle ‘select the second highest ranking argument in the logical structure as undergoer’. With a simple transitive verb, undergoer selection will work exactly the same way as in direct–indirect object languages, but when the verb has three arguments, this will always select the y argument as undergoer, never the z argument. The superiority of this second analysis can be seen clearly in the analysis of causative constructions in primary object languages. Yaqui (Uto-Aztecan) is a primary object language (Félix 2000), and the following examples from Guerrero Valenzuela and Van Valin (2004) illustrate three-place verbs and causatives. (4.59) a.

ʔ uka vaci-ta miika-k. Joan-Ø Peo-ta Juan-NOM Pedro-ACC DET.ACC corn-ACC give-PERF ‘Juan gave Pedro the corn.’

b.

ʔ uka vaci-ta miik-wa-k. Peo-Ø Pedro-NOM DET.ACC corn-ACC give-PASS-PERF ‘Pedro was given the corn.’

b .

*Uʔ u vaci-Ø Peo-ta miik-wa-k. DET.NOM corn-NOM Pedro give-PASS-PERF ‘The corn was given Pedro.’

In (4.59a) both non-privileged syntactic argument NPs are in the accusative case; which one is the undergoer and which one is the non-macrorole direct core argument? The answer is given by passive; in Yaqui only the undergoer can be privileged syntactic argument in a passive, and the actor is obligatorily omitted. Of the two logically possible passive versions of (4.59a), only (4.59b) with the recipient as privileged syntactic argument is grammatical. Hence the recipient Peo ‘Pedro’ is the undergoer in (4.59a). The facts in (4.59) could be accounted for by either of the analyses given above. A causative construction involving miika- ‘give’ presents an interesting problem, since there will be at least two non-privileged syntactic argument animate NPs; which one is the undergoer of the derived causative verb miik-tua ‘cause to give’? Again, passive provides the crucial evidence. (4.60) a.

Uʔ u maejto usi-ta mansana-ta yoem-ta miik-tua-k. DET.NOM teacher child-ACC apple-ACC man-ACC give-CAUS-PERF ‘The teacher made the child give the man the apple.’

b.

usi-Ø mansana-ta yoem-ta miik-tua-wa-k. Uʔ u DET.NOM child-NOM apple-ACC man-ACC give-CAUS-PASS-PERF ‘The child was made to give the man the apple.’

b .

*Yoem-Ø usi-ta mansana-ta miik-tua-wa-k. man-NOM child-ACC apple-ACC give-CAUS-PASS-PERF


b . c.

*Mansana-Ø usi-ta yoem-ta miik-tua-wa-k. apple-NOM child-ACC man-ACC give-CAUS-PASS-PERF [do (maejto, Ø)] CAUSE [[do (usi, Ø)] CAUSE [BECOME have (yoem, mansana)]]

The only grammatical version of (4.60a) is (4.60b) with the causee usi ‘child’ as privileged syntactic argument, and this shows that the undergoer in (4.60a) is usi-ta ‘the child-ACC’. With respect to the logical structure of (4.60a) in (4.60c), the animate-as-undergoer principle does not uniquely pick out usi ‘child’ as the undergoer, although it is compatible with it. However, the ‘second highest argument in logical structure’ principle for selecting the undergoer does correctly identify usi ‘child’ as the unique undergoer choice, and therefore it appears to be the correct characterization of undergoer selection in primary object languages. The situation in Yaqui is in fact more complex and hence even more interesting. There are verbs, such as nenka ‘sell’, which lexicalize the English pattern. This is exemplified in (4.61). (4.61) a.

Joan-Ø Peo-ta-u ju-ka kaba’i-ta nenka-k. Juan-NOM Pedro-ACC-DIR DET-ACC horse-ACC sell-PERF ‘Juan sold the horse to Pedro.’

b.

Ju-Ø kaba’i-Ø Peo-ta-u nanka-wa-k. DET-NOM horse-NOM Pedro-ACC-DIR sell-PASS-PERF ‘The horse was sold to Pedro.’

b .

*Peo-Ø ju-ka kaba’i-ta nenka-wa-k. Pedro-NOM DET-ACC horse-ACC sell-PASS-PERF ‘Pedro was sold the horse.’

With nenka ‘sell’, the theme is the undergoer and the recipient is marked by the postposition u ‘to’. Hence in a passive, the undergoer-theme appears as the privileged syntactic argument, rather than the undergoer-recipient as in (4.59). Moreover, there are verbs in Yaqui which exhibit the locative alternation; one of them is chijakta- ‘sprinkle’, illustrated in (4.62). (The data are from Guerrero and Van Valin 2004.) (4.62) a.

a .

125

Empo kafe-ta mesa-po chijakta-k. 2sgNOM coffee-ACC table-LOC sprinkle-PERF ‘You sprinkled coffee on the table.’ Kafe-Ø mesa-po chijakta-wa-k. coffee-NOM table-LOC sprinkle-PASS-PERF ‘Coffee was sprinkled on the table.’

a .

*Mesa-Ø kafe-ta chijakta-wa-k. ‘The table was sprinkled coffee.’

b.

Empo kafe-ae mesa-ta chijakta-k. 2sgNOM coffee-INST table-ACC sprinkle-PERF ‘You sprinkled the table with coffee.’

126


b .

Mesa-Ø kafe-ae chijakta-wa-k. table-NOM coffee-INST sprinkle-PASS-PERF ‘The table was sprinkled with coffee.’

b .

*Kafe mesa-ta chijakta-wa-k. ‘Coffee was sprinkled table.’

c.

[do (2sg, [sprinkle (2sg, kafe)])] CAUSE [BECOME be-on (mesa, kafe)]

The pattern here is the same as in the English examples in (4.40b,b ). In (4.62a), the lowest ranking argument in the logical structure in (4.62c), kafe ‘coffee’, is selected as undergoer, and the other argument is marked by the locative postposition –po. It is the undergoer argument that appears in the nominative case in the passive in (a ). In the alternative form in (b), the location argument, mesa ‘table’, is selected as undergoer; it appears in the accusative case, and the conditions for the instrumental case rule are met, resulting in kafe ‘coffee’ being marked by the instrumental case. What these examples show is that there are three different patterns in Yaqui: the direct–indirect object pattern in (4.61), the primary object pattern in (4.59) and the locative alternation pattern in (4.62). Yaqui is considered to be a primary object language, because the pattern in (4.59) is found with many ditransitive verbs and is the pattern found in all derived verb forms, such as the causative verbs in (4.60). This analysis has important implications for the actor–undergoer hierarchy: the principle governing the selection of the undergoer argument is different in primary object languages from direct object languages, and consequently the markedness relations expressed in Figure 2.4 are not true universally. Accordingly the hierarchy must be stated as in Figure 4.4. The actor selection principle is the same as before. When the verb has only two arguments, then the two undergoer selection principles are equivalent and always pick out the same argument as undergoer. However, when the verb has three arguments, then the difference between the two principles comes into play, yielding the different patterns with ditransitive verbs seen above. English dative shift could be viewed either as a marked undergoer selection in terms of Principle A or as the application of Principle B. In Yaqui clearly both principles are operative: some lexical verbs and all derived verbs follow Principle B, some verbs take Principle A as an absolute, not as a default, e.g. nenka ‘sell’, while others take Principle A as a default, e.g. chijakta‘sprinkle’. ACTOR Arg. of DO

UNDERGOER 1st arg. of 1st arg. of do′ (x,... pred′ (x, y)

2nd arg. of pred′ (x, y)

Arg. of pred′ (x)

Actor selection: highest ranking argument in LS Undergoer selection: Principle A: lowest ranking argument in LS (default) Principle B: second highest ranking argument in LS

Figure 4.4 Actor–undergoer hierarchy (revised)


It should be noted that this contrast in undergoer selection principles is independent of the privileged syntactic argument selection principles of the language, as is to be expected. Both types are found in syntactically accusative and syntactically ergative languages. English and Yaqui are both syntactically accusative, while Dyirbal (an indirect object language) and Belhare (a secondary object language; Bickel 2002) are syntactically ergative.21 21

Given this typological fact, it is perhaps better to refer to these two types of languages as ‘indirect object’ and ‘secondary object’ languages, avoiding the potential problem of saying that, e.g., Dyirbal is a syntactically ergative, direct object language. See Guerrero and Van Valin (2004) for more detailed discussion.

127

5

Linking syntactic and semantic representations in simple sentences

5.0


The various components of a description of grammatical structure have been presented (clause structure, lexical representation and semantic roles, syntactic functions, focus structure), and now the principles that link them together will be presented. They illustrate the workings of the syntax–semantics– pragmatics interface. The linking algorithm is central to a theory like RRG that posits only one level of syntactic representation, for it must be able to deal not only with canonical clause patterns, i.e. those in which the default correlations between syntactic and semantic structure exist, but also with the non-canonical patterns as well. The general linking schema in RRG may be sketched as in Figure 5.1 on p. 129. The relation between logical structure and macroroles is mediated by the actor–undergoer hierarchy in Figure 4.4. The relation between macroroles (and non-macrorole arguments of the verb) and morphosyntactic functions is subject to extensive cross-linguistic variation and is affected by the privileged syntactic argument selection hierarchy in (4.14) and selection principles in (4.15) and by the extent to which focus structure is grammaticalized in clause-internal relational syntax (see Figure 4.3). The opposition labelled ‘universal’ vs ‘language-specific’ in Figure 5.1 reflects the fact that there is very little cross-linguistic variation in the lexical phase of the linking and a great deal of cross-linguistic variation in the syntactic phase. The primary variation in the lexical phase is limited to three areas: what role animacy plays in macrorole assignment (e.g. in Lakhota, actor NPs must be animate), whether a language allows variable undergoer selection, and whether the language follows the ‘indirect object’ pattern of lowest-ranking argument in logical structure as undergoer or the ‘secondary object’ pattern of second-highest ranking argument in logical structure as undergoer. One of the most intriguing findings in Van Valin and LaPolla (1997) is a correlation between semantic motivation and universality: the more semantically motivated or semantically driven a grammatical phenomenon is, the less likely it is to show cross-linguistic variation (see also Van Valin 2003). An example of this has already been discussed in chapter 1: the units of the layered structure of the clause are semantically motivated and universal, while other constituents such as the left-detached position and the precore slot are pragmatically rather than semantically motivated and are not universal.

128

SYNTACTIC FUNCTIONS: PSA

Direct core arguments

Privileged syntactic argument [PSA] selection: Highest ranking MR = default (e.g. English) Lowest ranking MR = default (e.g. Dyirbal) SEMANTIC MACROROLES: ACTOR Arg. of 1st arg. of 1st arg. of DO do′ (x,...) pred′ (x,y)

Oblique core arguments

2nd arg. of pred′ (x,y)

Languagespecific

5.1 The linking algorithm

UNDERGOER Arg. of state pred′ (x)

Verb class

Logical structure

STATE predicate′ (x) or (x, y) ACTIVITY do′ (x, [predicate′ (x) or (x, y)]) ACHIEVEMENT INGR predicate′ (x) or (x, y) SEMELFACTIVE SEML predicate′ (x) or (x, y) ACCOMPLISHMENT BECOME predicate′ (x) or (x, y) ACTIVE ACCOMPLISHMENT do′ (x, [predicate 1′ (x, (y))]) & INGR predicate 2′ (z, x) or (y) α CAUSE β, where α , β are LSs of any type CAUSATIVE

Universal

Transitivity = No. of macroroles [MR α ] Transitive = 2 Intransitive = 1 Atransitive = 0 Argument positions in LOGICAL STRUCTURE

Figure 5.1 Summary of RRG linking system

5.1

The linking algorithm

A distinctive feature of the RRG linking algorithm is that it is bidirectional; that is, it links the semantic representation to the syntactic representation, and it also links the syntactic representation to the semantic representation. Viewed in terms of a language processing model, the semantics-to-syntax linking is an aspect of the production process, while the syntax-to-semantics linking is an aspect of the comprehension process. In the comprehension process, the parser would take the input and produce a structured syntactic representation of it, identifying the elements of the layered structure of the clause and the cases, adpositions and other grammatically relevant elements in the sentence. It is then the task of the grammar to map this structure into a semantic representation, as the first step in interpreting it, and this is where the syntax-to-semantics linking algorithm is required. The same syntactic and semantic representations are used in both linking algorithms.1 The linking between semantic and syntactic representations is governed by a very general constraint, the completeness constraint, which is stated in (5.1). (5.1)

1

Completeness constraint: All of the arguments explicitly specified in the semantic representation of a sentence must be realized syntactically in the sentence, and all of the

For discussions of RRG in relation to theories of sentence processing, both production and comprehension, see Butler (2004) and Van Valin (2005).

129

130

linking in s imple sentences

referring expressions in the syntactic representation of a sentence must be linked to an argument position in a logical structure in the semantic representation of the sentence.2

In simple sentences, this guarantees that there will be a match between the number of arguments in the clause and in the logical structure of the verb. As will be seen in chapter 7, it is extremely important in linking in complex sentence constructions. The semantic representation of a sentence is built around the logical structure of the predicator, usually a verb, and it is put together in the lexicon. For the semantics-to-syntax linking, the information in the semantic representation is crucial for the selection of the syntactic template(s) constituting the syntactic representation. In chapter 1, it was suggested that syntactic representations were best conceived of as ‘syntactic templates’, stored in what is called the ‘syntactic inventory’. There are principles governing the selection of the appropriate core template; they are given in (5.2). (5.2) a.

Syntactic template selection principle: The number of syntactic slots for arguments and argument-adjuncts within the core is equal to the number of distinct specified argument positions in the semantic representation of the core.

b.

Language-specific qualifications of the principle in (a): 1. All cores in the language have a minimum syntactic valence of 1. 2. Argument-modulation voice constructions reduce the number of core slots by 1. 3. The occurrence of a syntactic argument in the pre/postcore slot reduces the number of core slots by 1 (may override (1) above).

The default principle in (5.2a) is straightforward: if a verb takes n arguments, then there needs to be n positions in the core for those arguments to appear in, in order to satisfy the completeness constraint in (5.1). The exceptions in (b) are language-specific: all of them apply to English, which requires dummy subjects for argumentless verbs like rain, which has a passive and in which WH-words occur in the precore slot, while none of them apply to Lakhota, for example. An example of an English syntactic template for a simple sentence with a transitive verb and one for an NP are given in Figure 5.2. Following the suggestion 2

‘Explicitly specified’ means that the argument position in the logical structure is filled by a variable or a constant; for it to be unspecified, it would be filled by ‘Ø’, as in (i). (i) a. b. b .

Max loaded the minivan. [do (Max, Ø)] CAUSE [BECOME be-in (minivan, Ø)] [do (Max, Ø)] CAUSE [BECOME be-in (minivan, the boxes)]

The logical structure in (ib) does not specify what the second argument of be-in is, unlike the one in (ib ), and it is not specified in the sentence in (ia). Hence only (ib) is a legitimate logical structure for (ia), because of the completeness constraint.


SENTENCE CLAUSE NP

CORE

COREN NUC

NUCN

PRED

N

V N V NUC

CORE

CLAUSE SENTENCE

Aspect Negation Directionals Directionals Modality Negation Status Tense Illocutionary Force

NASP

NUC N

NUM QNT NEG

COREN

DEF DEIC

NP

Figure 5.2 Syntactic templates for a simple clause and an NP in English

made in section 1.3, the core template will not carry categorial specifications for its positions. The operators that will be needed in a particular sentence or NP are a function of its semantic representation (cf. sections 2.2, 2.3). Peripheries are not specified in the templates, as they are always optional. For syntax-to-semantics linking, the syntactic representation is created by the parser, as stated earlier. These considerations lead to a revised picture of the organization of RRG; it is presented in Figure 5.3.

Parser

Lexicon

Linking algorithm

S EMANTIC REPRESENTATION

Discourse-pragmatics

SYNTACTIC REPRESENTATION

Syntactic inventory

Figure 5.3 Organization of Role and Reference Grammar (revised)

5.1.1

Constructional schemas

RRG recognizes the importance of grammatical constructions, and they are represented in terms of constructional schemas. Cross-constructional and cross-linguistic generalizations are captured in terms of the general principles

131

132


and constraints that constitute the linking algorithms, e.g. the actor–undergoer hierarchy, the layered structure of the clause, the privileged syntactic argument selection hierarchy. Only the idiosyncratic, language-specific features of constructions are represented in constructional schemas.3 Hence constructional schemas, by virtue of their reference to general principles, permit the capturing of crosslinguistic generalizations, while at the same time expressing language-particular properties of grammars. The English passive construction is a good example of interaction of the general and the language-specific. The general characterization of voice constructions in (4.43) defines the general features of a passive construction, but the specific features of the English passive are represented in the constructional schema in Table 5.1. Each constructional schema contains syntactic, morphological, semantic and pragmatic information about the construction in question. This schema expresses the specific features of the English passive. Table 5.1 Constructional schema for English passive (plain) Construction : English passive (plain) Syntax : Template(s): (5.2b2) PSA: (4.15a,c2), Variable [± pragmatic influence] Linking: (4.43a) (4.43b): omitted or in peripheral by-PP Morphology: Verb: past participle Auxiliary: be Semantics : PSA is not instigator of state of affairs but is affected by it (default) Pragm at i c s: Illocutionary force: unspecified Focus structure: no restrictions; PSA = topic (default)

As an example of cross-linguistic comparison, consider the English and Sama WH-question constructions. Sama is a syntactically ergative language, and both languages have a precore slot in which the WH-word appears. However, there is a restricted neutralization in Sama: the WH-word must be the S, UT or d-S of the clause. This is illustrated in (5.3). (5.3) a.

b.

3

Say nda’ d’nda? who see woman ‘Who did the woman see?’ (*‘Who saw a woman?’) Say N-nda’ d’nda? who ANTI-see woman ‘Who saw a woman?’ (*‘Who did the woman see?’)

These were termed ‘constructional templates’ in Van Valin and LaPolla (1997).


Table 5.2 Constructional schema for Sama antipassive Construction : Sama antipassive Syntax : Template(s): (5.2) PSA: (4.15b,c1), Variable [± pragmatic influence] = immediate postverbal direct core argument Linking: (4.43a) (4.43b): none if U is 3rd person; oblique core argument if U is non-3rd person Morphology: Verb: N- + verb stem Oblique core argument: ma + pronoun Semantics: PSA is instigator of state of affairs Aktionsart = atelic (default) Pragmatics: Illocutionary force: unspecified Focus structure: U = non-specific (default) PSA = topic (default)

When the WH-word is the undergoer, the plain form of the verb is used, as in (5.3a), but if it is the actor, then the antipassive form of the verb must be used, as in (b). The constructional schema for the Sama antipassive is given in Table 5.2. The fact that there is a restricted neutralization with respect to the WH-word in questions means that there is never any ambiguity as to the function of the WHword in the precore slot in Sama, unlike in English. The constructional schemas for English and Sama WH-question constructions are given in Tables 5.3 and 5.4. The syntactic template, semantics and pragmatics of the two constructions are the same; the crucial difference between them lies in the lack of a restricted Table 5.3 Constructional schema for English WH-questions Construction : English WH-question Syntax : Template(s): PrCS, (5.2a,b3) PSA: None Linking: WH-XP to PrCS Morphology: Default Semantics : contains an open proposition with a variable α, WH-XP = α Pragm at i c s: Illocutionary force: interrogative Focus structure: narrow focus on PrCS

133

134


Table 5.4 Constructional schema for Sama WH-question formation Construction : Sama WH-question formation Syntax : Template(s): PrCS, (5.2a,b3) PSA: (4.15b,c1), WH-NP = Variable pivot [+ pragmatic influence] Linking: WH-NP to PrCS Antipassive, if WH-NP = default choice for PSA Morphology : None Semantics : Contains an open proposition with variable α, WH-NP = α Pragm at i c s: Illocutionary force: interrogative Focus structure: narrow focus on PrCS

neutralization in the English construction and the existence of a variable syntactic pivot in the Sama construction. Constructional schemas play an important role in linking in both directions. This leads to a revision of the organization of RRG, which is presented in Figure 5.4. For the semantics-to-syntax linking they supply the language-specific and construction-specific details which are required for the correct encoding of meaning in the morphosyntax. They play a crucial role in the syntax-to-semantics linking, especially in languages which have different privileged syntactic arguments for different constructions (e.g. Jakaltek, Sama), because they specify what the privileged syntactic argument is in the construction, which is central to the linking. As an example, consider the interaction of WH-question formation and reflexivization in Sama. In this language, the actor is always the controller for reflexivization, regardless of whether the clause is active or antipassive voice. The schema for Sama reflexivization is given in Table 5.5. (The role hierarchy and l s -superiority conditions will be introduced in section 5.3.) Parser SYNTACTIC REPRESENTATION

Lexicon

Discourse-pragmatics

Syntactic inventory

Linking algorithm

Constructional schemas

SEMANTIC REPRESENTATION

Figure 5.4 Organization of Role and Reference Grammar (final)


Table 5.5 Constructional schema for Sama reflexivization Construction : Sama reflexivization Syntax : Template(s): (5.2a) PSA: semantic controller = actor Linking: Actor = direct core argument Morphology: Reflexive: di + genitive pronoun (person, number) Semantics : Actor and reflexive pronoun are obligatorily coreferential within the clause Subject to role hierarchy and l s-superiority conditions Pragm at i c s: Illocutionary force: unspecified Focus structure: unspecified

The interaction of these two constructions is illustrated in (5.4). (5.4) a.

B’lli d’nda daing ma di-na. buy woman fish for REFL-3sg ‘The woman bought the fish for herself.’

a .

Ay b’lli d’nda ma di-na? what buy woman for REFL-3sg ‘What did the woman buy for herself?’

b

N-b’lli d’nda daing ma di-na. ANTI-buy woman fish for REFL-3sg ‘The woman bought fish for herself.’

b .

Say n-b’lli daing ma di-na? who ANTI-buy fish for REFL-3sg ‘Who bought fish for him/herself?’

In (5.4a), d’nda ‘woman’, the actor NP, is the controller of the reflexive dina. This is also the case in the WH-question in (a ); in this sentence, the actor is the privileged syntactic argument (semantic controller) for reflexivization, and the undergoer is the privileged syntactic argument (syntactic pivot) for WH-question formation. The third example presents a clause in the antipassive voice, and the actor d’nda is the controller of the reflexive. In the final example, the privileged syntactic argument of the WH-question and the privileged syntactic argument of reflexivization are the same argument, the actor NP say ‘who’. Thus, the analysis of (5.4a ,b ) involves three constructional schemas: the one for antipassive voice in Table 5.2, the one for WH-questions in Table 5.4, and the one for reflexivization in Table 5.5. The syntax-to-semantics linking algorithm for Sama would refer to these schemas at crucial points to guide the interpretive process (see Figure 5.21).

135

136


5.1.2

From semantics to syntax

The linking procedure from semantics (logical structure) to syntax (layered structure of the clause) is summarized in (5.5).4 These are the general linking principles, which may be overridden by the specific requirements of a construction as expressed in its constructional schema. (5.5)

Linking algorithm: semantics → syntax 1. Construct the semantic representation of the sentence, based on the logical structure of the predicator. 2. Determine the actor and undergoer assignments, following the actor–undergoer hierarchy in Figure 4.4. 3. Determine the morphosyntactic coding of the arguments a. Select the privileged syntactic argument, based on the privileged syntactic argument selection hierarchy and principles in (4.14)–(4.15). b. Assign the arguments the appropriate case markers and/or adpositions. c. Assign the agreement marking to the main or auxiliary verb, as appropriate. 4. Select the syntactic template(s) for the sentence following the principles in (5.2). 5. Assign arguments to positions in the syntactic representation of the sentence. a. Assign the [−WH] argument(s) to the appropriate positions in the clause. b. If there is a [+WH] argument of a logical structure, 1. assign it to the normal position of a non-WH-argument with the same function, or 2. assign it to the precore or postcore slot, or 3. assign it to a position within the potential focus domain of the clause (default = the unmarked focus position). c. A non-WH argument may be assigned to the precore or postcore slot, subject to focus structure restrictions (optional). d. Assign the [−WH] argument(s) of logical structure(s) other than that of the predicator in the nucleus to 1. a periphery (default), or 2. the precore or postcore slot, or 3. the left- or right-detached position.

All steps are all subject to cross-linguistic variation. As discussed in section 4.5, some languages allow variable undergoer selection while others do not, and languages differ as to the principle governing undergoer selection (indirect object vs secondary object languages) (step 2). The privileged syntactic argument selection principles in (4.15) vary along two major parameters, accusative vs ergative privileged syntactic argument selection and whether privileged syntactic argument 4

The linking algorithm in (5.5) differs from the one presented in Van Valin and LaPolla (1997) in two ways. First, steps 1 and 4 are presupposed by the algorithm presented there, and they are made explicit here. Second, the order of the steps is somewhat different; the reasons for this change will be spelled out below.


selection is restricted to macrorole arguments or not (step 3a); information from constructional schemas can play a crucial role at this point. Case and agreement show substantial cross-linguistic variation (step 3b,c). The positions to which XPs are assigned in sentences varies not only within languages but across languages (step 5a), and the possibilities under step 5b cover the range of WH-question types found in human languages. Thus (5.5) lays out the general linking algorithm, which would have to be specialized for each individual language. The workings of (5.5) can be illustrated by going through it with respect to the realization of several related versions of a sentence. As an example, let’s take a speaker wanting to express that Sandy transferred some flowers to Chris at a party. There are a number of verbs that could be used here, and in this instance the speaker chooses present. The system in (5.5) presumes that a speaker is realizing a specific communicative intention, and consequently whether the sentence will be, e.g., active or passive, declarative or interrogative, figures into the formulation of the semantic representation and concomitant syntactic template selection. Moreover, the discourse status (activation level) of the referents of the NPs is also represented (see section 3.4). The output of step (1), which takes place in the lexicon (see section 5.2), is given in Figure 5.5. The operators for the NPs are not represented here, in the interest of space. 〈 IFDEC 〈 TNSPAST 〈be-at¢ (partyACS, [[do¢ (SandyACV, Ø)] CAUSE [BECOME have¢ (ChrisACS, flowersACV)]])〉〉〉

Figure 5.5 Output of step 1 in (5.5)

The next step is actor and undergoer assignment; this is illustrated in Figure 5.6, in which irrelevant details are not specified.5 This reflects the unmarked or default macrorole assignments for English: the leftmost argument, Sandy, is the actor, and the rightmost argument, the flowers, is the undergoer. This leaves the third NP, Chris, as a non-macrorole argument. The representation in Figure 5.6 is not a distinct level of representation from the one in Figure 5.5; rather, it is simply the same representation enriched with additional information, namely the ...[do¢ (ACT: SandyACV, Ø)] CAUSE [BECOME have¢ (NMR: ChrisACS, UND: flowersACV)]...


assignment of macroroles. The next step concerns the morphosyntactic properties of the arguments, specifically privileged syntactic argument selection and case/adposition assignment. The most important part of step (3) is privileged 5

When a single participant appears in more than one argument position in a logical structure, as in [do (x, Ø)] CAUSE [BECOME have (x, y)] ‘take’ or do (x, [eat (x, y)]) & [INGR consumed (y)] ‘eat’ (active accomplishment), its linking properties with respect to actor are determined by the higher ranking position and with respect to undergoer by the lower ranking position. So with take the x argument is both effector and recipient, and the linking rules treat it as an effector with respect to actor selection; similarly, with eat the y argument is both consumed and patient, and it is treated as a patient with respect to macrorole selection.

137

138


...[do¢ (ACT: SandyACV, Ø)] CAUSE [BECOME have¢ (NMR: to ChrisACS, UND: flowersACV)]... [PSA: NOM] Active, 3sg [ACC] [ACC]


syntactic argument selection, and in this instance the actor is chosen as privileged syntactic argument, yielding an active voice sentence (step 3a). This determines the case marking of the clause: Sandy, as the highest ranking core macrorole, would receive nominative, and the flowers, as the other core macrorole, would receive accusative. Chris is a non-macrorole core argument, and the conditions for the to-assignment rule in (4.39a) are met, resulting in Chris being assigned to. All prepositions in English assign accusative case to their objects; of course, since full NPs do not inflect for case in English, these rules apply vacuously. Their application would only be apparent if the argument variables in the logical structure were filled by pronouns. The verb is active voice, as noted above, and the result of step 3c is that it would show 3sg agreement; however, since only be shows person differences in the past tense, the agreement features would not be realized morphologically. The next step is syntactic template selection (step 4), and the result of this is shown in Figure 5.8. Proper nouns normally do not have a layered structure, hence the lack of operator projections in two of the NPs. The semantic representation of the NP the flowers would be DEF + NEG − QNT ∃ NUM PL NASP COUNT flower, and this determines the operators to be found in the operator SENTENCE CLAUSE CORE NUC PRED V V NUC PERIPHERY

CORE CLAUSE

Tense

NP

CLAUSE

IF

COREN NUCN

SENTENCE

N PP P

PP NP

NP

COREP NUCP

NPROP NP

N

PRED

NUCN

P

COREN DEF


NP

NUM


SENTENCE CLAUSE CORE NP N

PROP

PERIPHERY

NUC

COREN

PRED

PP

PP

NP P

NUCN

NP

COREP NP NUCP

N

PRED

PROP

V Sandy

P

N

presented

COREN NUCN

the flowers

to

Chris

at

the

N party.

N

N

NUC

NUC N

NUC N

CORE

COREN

V

CLAUSE CLAUSE

DEF

Tense

COREN

NUM

NP

DEF

NUM

NP

IF

SENTENCE

Figure 5.9 Result of all steps in (5.5)

projection in the syntactic representation of the NP; of these operators, only definiteness and number are realized overtly in the morphosyntax. The same would be true for the NP the party. The NPs would be linked to their respective NP templates. The final step involves linking the XPs into the structure in Figure 5.8; this is represented in Figure 5.9. The NP Sandy, as the privileged syntactic argument, is linked to the core-initial position, and the NP the flowers, as the undergoer, is linked to the immediately postverbal position. The PP to Chris follows the undergoer as the last phrase in the core. The fact that the order in the core must be NP V NP PP follows from independent, language-specific principles: in English, the privileged syntactic argument must be the first NP in the core, and the undergoer (when it is not the privileged syntactic argument) must immediately follow the nucleus. The predicative adjunct PP at the party is linked to the peripheryCORE . The verb present is linked to the nucleus and is inflected for past tense, and the core-internal position of the morpheme expressing the tense operator signals declarative illocutionary force. It will not be possible to give this much detail for every sentence that will be analysed, and consequently the steps illustrated in Figures 5.4 through 5.8 can be abbreviated as in Figure 5.10. In such abbreviated diagrams, the operator projections and the internal structure of NPs and PPs will not be specified unless it is relevant to the point under discussion. All of the relevant linking information from Figures 5.4–5.8 is summarized in Figure 5.10 on the next page, with the outputs from steps (2) and (3) depicted above the semantic representation of the sentence for clarity. The alternative undergoer choice is presented in Figure 5.11. In this sentence Chris is selected as undergoer, rather than the flowers. Because there is a marked

139

140

linking in s imple sentences 4

SYNTACTIC INVENTORY

SENTENCE CLAUSE CORE

PERIPHERY

NUC

NP

NP

PP

PP

PRED V Sandy

presented

the flowers

to Chris

at the party.

5 at: ACC

ACTIVE: 3sg

PSA:NOM

to: ACC

ACC

3

be-at¢ (party

LEXICON

NMR

ACTOR

1 ACS

2

UNDERGOER

, [[do¢ (SandyACV , Ø)] CAUSE [BECOME have¢ (ChrisACS, flowersACV)]])

Figure 5.10 Abbreviated linking diagram summarizing Figures 5.5–5.9

4

SYNTACTIC INVENTORY

SENTENCE CLAUSE CORE NUC

NP

PERIPHERY NP

PP

PP

PRED V Sandy

presented

Chris

with the flowers at the party.

5 at: ACC

ACTIVE: 3sg

PSA:NOM

ACC

with:ACC

3 ACTOR

1 LEXICON

UNDERGOER NMR

2

be-at¢ (party

ACS

, [[do¢ (Sandy

ACV

, Ø)] CAUSE [BECOME have¢ (Chris ACV, flowersACS )]])

Figure 5.11 Linking in sentence with marked undergoer choice with present

undergoer choice, the with assignment rule in (4.29a) applies, resulting in the NP the flowers being marked by with. These last two figures represent the two possible active voice linkings with present; one of the possible passive voice linkings is given in Figure 5.12. The macrorole assignments in Figure 5.12 are the same as in Figure 5.11, but in this figure the undergoer rather than the actor is selected as the privileged syntactic argument, yielding a passive construction. Accordingly, the actor NP Sandy is assigned the preposition by and is linked to a position in the peripheryCORE , while the undergoer NP Chris is linked to the core-initial privileged syntactic argument position, following the constructional schema in Table 5.1.


4

SYNTACTIC INVENTORY

SENTENCE CLAUSE CORE NP

PERIPHERY PP

NUC PRED

AUX

PP

PP

V Chris

was presented

with the flowers by Sandy

at the party.

5 PASSIVE: 3sg

by: ACC

at: ACC

PSA:NOM with:ACC

3 ACTOR

1 be-at¢ (party

LEXICON

ACS

UNDERGOER NMR

2 , [[do¢ (Sandy INA , Ø)] CAUSE [BECOME have¢ (ChrisACV, flowersACV)]])

Figure 5.12 Passive voice linking of sentence in Figure 5.11

WH-questions have long been an important topic in syntactic theory, especially the kind found in English which involve a long-distance dependency. A phenomenon which all syntactic theories must account for is the case which may appear on a WH-word in the precore slot, as in, e.g., Whom did Sandy present with the flowers at the party? As Figure 5.13 shows, this follows without any further specifications from the RRG linking algorithm. It should be noted that simple WH-words like who and what have no activation level, since in the question they have no specific referent. WH-NPs like which boy would have one,

SYNTACTIC INVENTORY

4


PrCS NP NP

PERIPHERY PP

PP

NUC PRED V

Whom

did

Sandy

present

with the flowers at the party?

5 at: ACC

PSA:NOM

ACC

ACTIVE: 3sg

with:ACC

3 ACTOR

1 LEXICON

UNDERGOER NMR

2

be-at¢ (party

ACS

, [[do¢ (SandyACV , Ø)] CAUSE [BECOME have¢ (who- , flowersACV)]])

Figure 5.13 Linking in a WH-question in English

141

142


since they presuppose that some set of boys has been previously mentioned or is accessible; such WH-expressions have often been referred to as being ‘discourse linked’. Since the WH-word who is the undergoer of present in an active voice construction, it would be assigned accusative case (at least in the grammar of the speakers who have whom) in step (3), and then in step (5) it would be linked directly to the precore slot, following the constructional schema in Table 5.3. In a complete representation as in Figure 5.9, did would be represented in the operator projection as expressing past tense and interrogative illocutionary force (by virtue of its core-initial position). English exhibits the rare phenomenon of preposition stranding in extraction constructions, e.g. Who did Sandy present the flowers to at the party?6 The semantics-to-syntax linking algorithm proposed in Van Valin and LaPolla (1997) cannot account for preposition stranding, because it orders the preposition assignment rule after the rule linking the XPs into the syntactic representation. Since the preposition can be linked independently of its NP object, it must be assigned before the linking of the NP to the syntax occurs. Since many contemporary speakers of English do not have whom in their grammar, the only possible form of a WH-question in which the WH-word is the object of a preposition is the stranded form. The linking in Who did Sandy present the flowers to at the party? is given in Figure 5.14. SYNTACTIC INVENTORY

4


PrCS NP NP

PERIPHERY

NUC

NP

PP

PP

PRED

P

V Who

did

Sandy

present

the flowers

to

at the party?

5 at: ACC

PSA:NOM

to: (ACC)

ACTIVE: 3sg

ACC

3 ACTOR

1 LEXICON

NMR UNDERGOER

2 be-at´ (partyACS , [[do´ (SandyACV , Ø)] CAUSE [BECOME have´ (who- , flowersACV )]])

Figure 5.14 Linking in WH-question with preposition stranding in English

There are two distinctive attributes of this construction. First, the preposition links to the third argument position in the core, just as in Figure 5.10, but it lacks an object. This is not unprecedented in English, because there are intransitive prepositions that can fill core slots, e.g. away as in He put the book away. Second, this structure violates the template selection principle in (5.2b3); the latter permits 6

See section 5.4.3 for a discussion of the discourse-pragmatic conditions on preposition stranding.


143

only To whom did Sandy present the flowers at the party?, since it requires that the core have one less argument slot to compensate for the occurrence of the argument XP in the precore slot. However, in an adjunct WH-question, e.g. Where did Sandy present the flowers to Chris?, there is a precore slot together with a full three-slot core, and accordingly this structure has a precedent in English syntax. Given the cross-linguistic rarity of this construction, this violation of (5.2b3) is normally not permitted by languages in which WH-words occur in the precore slot. The accusative case assigned by to is in parenthesis, since it is optional (and dispreferred) in this construction even for speakers who have whom in their grammar. If the question were Where did Sandy present Chris with the flowers?, then the logical structure would be be-loc (where, [[do (SandyACV , Ø)] CAUSE [BECOME have (ChrisACV , flowersACV )]]). The core internal linking would be the same as in Figure 5.11, and step 5b2 in (5.5) would assign the adjunct WHword where to the precore slot. It would be useful to illustrate the steps in (5.5) with respect to a language with a full case system, e.g. German. Consider the examples in (5.6); their semantic representation is given in (5.6d). (5.6) a.

b.

c.

d.

Der Mann ha-t der Frau den Hut geschenkt. the.MsgNOM man have-3sgPRES the.FsgDAT woman the.MsgACC hat give.PSTP ‘The man gave the hat to the woman [as a gift].’ We-m ha-t der Mann den Hut geschenkt? who-DAT have-3sgPRES the.MsgNOM man the.MsgACC hat give.PSTP ‘To whom did the man give the hat?’ Den Hut ha-t der Mann der Frau geschenkt. the.MsgACC hat have-3sgPRES the.MsgNOM man the.FsgDAT woman give.PSTP ‘The hat the man gave to the woman.’ IF DEC/INT TNS PRESASP PERF [do (Mann, Ø)] CAUSE [BECOME have (Frau/we-, Hut)]

The linking in (5.6b), which has a WH-word in the precore slot, is presented in Figure 5.15, while the linking in (5.6c), which has a non-WH NP in the precore slot, is given in Figure 5.16 on p. 144. The output of step (2), macrorole assignment, is always the same, as German does not allow variable linking to undergoer like English; hence Hut ‘hat’ is always the undergoer and Frau ‘woman’ or we‘who’ is always a non-macrorole core argument. In step (3) the actor is selected as privileged syntactic argument, yielding an active construction. The actor NP, Mann ‘man’, is the highest ranking core macrorole, and it therefore receives nominative case, and the finite verb would agree with it. The undergoer NP, Hut ‘hat’, is the other macrorole, and therefore it receives accusative case, while the non-macrorole core argument, Frau ‘woman’ or we- ‘who’, receives dative case. In both of these constructions, the NP in the precore slot is linked directly from its logical structure position to the precore slot, and no special considerations are required to account for its case.

144


SENTENCE 4

SYNTACTIC INVENTORY

CLAUSE CORE PrCS

NP

NP

NUC PRED

NP Wem

V hat

der Mann

den Hut geschenkt?

5 PSA:NOM

ACTIVE: 3sg

DAT

ACC

3 ACTOR

1

NMR UNDERGOER

2 [do¢ (Mann ACV, Ø)] CAUSE [BECOME have¢ (we-, Hut ACV)]

LEXICON

Figure 5.15 Linking from semantics to syntax in (5.6b) from German

SENTENCE 4

SYNTACTIC INVENTORY

CLAUSE CORE PrCS

NP

NP

NUC PRED

NP Den Hut

V hat der Mann

der Frau geschenkt.

5 PSA:NOM

ACTIVE: 3sg

DAT

ACC

3 1 LEXICON

ACTOR

2

NMR

UNDERGOER

[do¢ (MannACV, Ø)] CAUSE [BECOME have¢ (Frau ACV , Hut INA )] Figure 5.16 Linking from semantics to syntax in (5.6c) from German

Variable undergoer selection is unusual for languages with case systems, but Dyirbal exemplifies both an ergative syntactic system and variable undergoer selection, as discussed in section 4.4. In Dyirbal, as in English and German, the rightmost argument in the logical structure is the default choice for undergoer. The relevant examples are given in (5.7), repeated from (4.35).


(5.7) a.

b.

c.

baŋgun d¸ugumbi-u wuga-n bagul yaa-gu. Balam mira-Ø NM.ABS beans-ABS NM.ERG woman-ERG give-TNS NM.DAT man-DAT ‘The woman gave beans to the man.’ Bayi yaa-Ø wuga-n baŋgun d¸ugumbi-u baŋgum mira-d¸u. NM.ABS man-ABS give-TNS NM.ERG woman-ERG NM.INST beans-INST ‘The woman gave the man beans.’ IF DECTNS NFUT[do (d¸ugumbil-, Ø)] CAUSE [BECOME have (yaa-, mira-)]

In Dyirbal the rules in (4.26) apply, assigning absolutive case to the undergoer, the lowest ranking macrorole, and ergative case to the actor d¸ugumbil ‘woman’.7 As in English, the choice of undergoer affects the coding of the third core argument. When the default selection is made, i.e. mira- ‘beans’ as undergoer, then the rule assigning dative case in (4.29b) applies, yielding (5.7a). When the marked selection is made, i.e. yaa- ‘man’, then instrumental rather than dative case is assigned to the non-macrorole core argument, following the rule in (4.29a), yielding (5.7b). The pattern in (5.7b) is analogous to that in Figure 5.11. The linking in (5.7b) is illustrated in Figure 5.17. SENTENCE

4

SYNTACTIC INVENTORY

CLAUSE CORE NP NUC

NP

NP

PRED V ¸ Bayi yaa wugan baŋgun dugumbiu

¸ baŋgum miradu.

5 ERG

ACTIVE

PSA: ABS

INST

UNDERGOER

NMR

3 1 LEXICON

ACTOR

[do¢ (dugumbil¸

2 ACS,

Ø)] CAUSE [BECOME have¢ (yaa -ACV , mira-INA )]

Figure 5.17 Linking from semantics to syntax in (5.7b) from Dyirbal

A different kind of variable undergoer assignment is found in possessor raising constructions like the one found in Acehnese in (4.4), repeated below. (5.8) a.

b.

7

145

Seunang até lôn. happy liver 1sg ‘I am happy.’ (lit: ‘My liver is happy.’) Lôn seunang-até. 1sg happy-liver ‘I am happy.’

Dyirbal has a person-based split case marking system; first and second person pronouns are subject to the rules in (4.25). See Van Valin and LaPolla (1997), section 7.3.1.2, for a discussion of splitergative case systems.

146


The logical structure for both of these sentences is be ([have.as.part (1sg, até)], [happy ]), and they differ with respect to undergoer selection. In the unmarked até )], [happy ]), linking in (5.8a), the logical structure is be ([have.as.part (lôn, ---and até lôn ‘my liver’ is selected as undergoer and is linked to the postnuclear core position. In the marked linking in (5.8b), the logical structure is be ([have.as.part (----lôn, até)], [happy ]), in which the possessor lôn is selected as head and hence as the undergoer. This is analogous to the alternation discussed in section 2.3 between the woman’s book (have (woman, book)) ------ and the woman woman, book)), in which either the possessed N (unmarked) with the book (have (---------or the possessor (marked) may be selected as the head of an NP expressing possession. The undergoer is linked to a core argument position, while the possessed body part is incorporated into the nucleus. These two linkings are illustrated in Figure 5.18. SENTENCE CLAUSE

SYNTACTIC INVENTORY

CORE 4

NUC

NP

PRED V Seunang

até lôn.

5 1

2 UNDERGOER be¢ ([have.as.part¢ (1sg, até)], [happy¢])

LEXICON

SENTENCE CLAUSE

SYNTACTIC INVENTORY

CORE

4

NUC

NP

PRED N V Lôn

seunang-até

5 1 LEXICON

2 UNDERGOER be¢ ([have.as.part¢ (1sg, até)], [happy¢])

Figure 5.18 Linking from semantics to syntax in Acehnese possessor raising construction


Head-marking languages present an additional complication to step 5a in (5.5), because when there are independent NPs an argument is in effect coded twice, once as a pronominal affix on the verb and once as an independent NP. The following two Lakhota examples illustrate this issue. (5.9) a.

b.

Wiˇchá-wa-kte. 3plU-1sgA-kill ‘I killed them.’ Mathó ki wiˇchá-wa-kte. bear the 3plU-1sgA-kill ‘I killed the bears.’

In (5.9a), which has no independent NPs, the linking is straightforward, especially since Lakhota is an [S, AT ] system and lacks a voice opposition. The linking in (5.9a) is given in Figure 5.19. The distinctive property of the clauses with one or more independent NPs is that both the independent NP and the bound argument marker on the verb fill the same argument position in the logical structure; in (5.9b) both mathó ki ‘the bear’ and wiˇcha- ‘3plU’ instantiate a single argument and necessarily have the same reference. The bound pronoun, the obligatory part, is linked to its slot in the verb, while the NP is linked to a position inside the clause but outside of the core. This is shown in Figure 5.20 on p. 148. SENTENCE SYNTACTIC INVENTORY

4

CLAUSE CORE PRO

PRO NUC PRED

Wicháwaˇ

5 PSA:NOM

1 LEXICON

V kte

ACC

3

ACTOR

UNDERGOER

2 [do¢ (1sg ACV , Ø)] CAUSE [BECOME dead¢ (3pl ACV )]

Figure 5.19 Linking from semantics to syntax in (5.9a) from Lakhota

In section 5.1.1, it was stated that constructional schemas play an important role in linking. The interaction of schemas was discussed with respect to the analysis of (5.4b ), which involves WH-question formation, antipassivization and reflexivization. This interaction can now be represented graphically, as in Figure 5.21 on p. 148. The schema for Sama WH-question formation in Table 5.4 specifies that (1) the WH-NP occurs in the precore slot, (2) the WH-NP must be the privileged syntactic argument in the clause, and (3) antipassive is required if the WH-NP does not function as the undergoer of the clause. Because the

147

148


4

SYNTACTIC INVENTORY

SENTENCE CLAUSE CORE PRO

NP

PRO NUC PRED V

Matho ki

5

wicháwaˇ

kte.

PSA:NOM

1

ACTOR

[do¢ (1sg

LEXICON

ACV

ACC

3

UNDERGOER

2 , Ø)] CAUSE [BECOME dead¢ (3pl [mathó]ACS )]

Figure 5.20 Linking from semantics to syntax in (5.9b) from Lakhota

SYNTACTIC INVENTORY

SENTENCE

4

CLAUSE CORE

PrCS NUC

WH-Q schema Linking to PrCS WH-NP = PSA Antipassive voice because WH-word ≠ Undergoer (Table 5.4)

NP

PRED

Antipassive schema Actor = PSA Undergoer = DCA (Table 5.2)

NP

V

Say

n-b’lli

PP daing

ma di-na? 5

PSA ACTOR

1 LEXICON

PERIPHERY

DCA

ANTI

3 2

ma: NP

UNDERGOER

[do¢ (say , [ buy¢ (say, daing)])] PURPa [BECOME have¢ (di-nai, daing)] i

Reflexive schema determines coindexing (Table 5.5) a ‘PURP’ is an abbreviation for the purposive logical structure, want (x, LS2 ) ∧ DO (x, [[LS1 ♦ CAUSE[LS2 ]]). See Van Valin and LaPolla (1997: 382–4) for discussion. Figure 5.21 Interaction of constructional schemas in the semantics-to-syntax linking in (5.4b )


WH-NP is in fact the actor, the antipassive schema in Table 5.2 is invoked. Finally, the reflexive binding is governed by the specifications in the schema in Table 5.5.

5.1.3

From syntax to semantics

Of the two directions of linking, going from the syntactic representation to the semantic representation is the more difficult, because it involves interpreting the overt morphosyntactic form of a sentence and deducing the semantic functions of the elements in the sentence from it. Accordingly, the linking rules must refer to the morphosyntactic features of the sentence. One of the main reasons there are so many steps in (5.10) is that it is intended to cover a significant cross-linguistic range of grammatical phenomena; the linking algorithm for any particular language will contain only those steps that are relevant to that language. The procedure for taking a sentence and linking it to its semantic representation is summarized in (5.10). (5.10)

Linking algorithm: syntax → semantics 1. Determine the macrorole(s) and other core argument(s) in the clause. a. If the verb is intransitive, then assign the privileged syntactic argument either macrorole or direct core argument status, depending upon the language (language-specific). b. If the verb is transitive and the language lacks voice oppositions, determine the macroroles from case marking and/or word order (language-specific). c. If the language has a voice opposition, determine the voice of a transitive verb (language-specific): 1. If the construction is syntactically accusative: a. If it is the unmarked voice, the privileged syntactic argument is actor. b. If it is passive, the privileged syntactic argument is not the actor of the predicate in the nucleus; 1. the actor may appear as a direct core argument (language-specific); or 2. the actor may appear in the peripheryCORE marked by an adposition or an oblique case (language-specific); or 3. if there is no actor in the core or the periphery, then replace the variable representing the highest ranking argument in the logical structure with ‘Ø’. 2. If the construction is syntactically ergative: a. If it is the unmarked voice, the privileged syntactic argument is undergoer. b. If it is antipassive, the privileged syntactic argument is actor; 1. the undergoer may appear as a direct core argument or as an oblique element (language-specific);

149

150


2.

3. 4.

5.

2. if there is no undergoer in the core or the peripheryCORE , then replace the variable representing the lowest ranking argument in the logical structure with ‘Ø’. 3. Assign macrorole status to the other direct core argument, if it is not dative or in an oblique case (language-specific). d. If the language is head-marking and there are independent NPs in the clause, associate each NP with a bound argument marker (language-specific). Retrieve from the lexicon the logical structure of the predicate in the nucleus of the clause and with respect to it execute step 2 from (5.5), subject to the following proviso: a. If the language allows variable undergoer selection and if there is more than one choice for undergoer, do not assign undergoer to an argument in the logical structure. b. Determine the linking of the non-macrorole core argument: 1. If there is a two-place state predicate in the logical structure and if the non-macrorole core argument is marked by a locative adposition or dative or a locative-type case, then link it with the first argument position in the state predicate in the logical structure and link the other non-actor core argument (if there is one) to the second argument position in the state predicate, or 2. If there is a two-place state predicate in the logical structure and if the non-macrorole core argument is not marked by a locative adposition or dative or a locative-type case, then link it with the second argument position in the state predicate and link the other non-actor core argument (if there is one) to the first argument position in the state predicate. 3. Otherwise, link the animate NP with the first argument position in the state predicate in the logical structure.8 Link the arguments determined in step 1 with the arguments determined in step 2 until all core arguments are linked. If there is a predicative adpositional adjunct, then retrieve its logical structure from the lexicon, insert the logical structure of the core as the second argument in the logical structure and the object of the adposition in the periphery as the first argument. If there is an element in the pre- or postcore slot (language-specific), a. Assign it the remaining unlinked argument position in the semantic representation of the sentence. b. And if there are no unlinked argument positions in the sentence, then treat the WH-word like a predicative preposition and follow the procedure in step 4, linking the WH-word to the first argument position in the logical structure.

The steps in 1c refer crucially to constructional schemas, especially in languages which have both syntactically ergative and syntactically accusative constructions. 8

This handles sentences like (4.48a), in which there are two non-actor core arguments in the same case.


The application of this linking algorithm to English can be illustrated with a sentence like Kim smashed the glass. The first step is to identify the verb and its voice: smashed is transitive and active voice, which means that the privileged syntactic argument is the actor. The NP following the verb is direct, and therefore it must be the undergoer. Step 2 involves retrieving the logical structure of smash from the lexicon, [do (x, Ø)] CAUSE [BECOME smashed (y)], and assigning macroroles, yielding ‘x = actor’ and ‘y = undergoer’. In step 3, the arguments from the sentence are linked to the logical structure arguments, yielding the result that ‘x = Kim’ and ‘y = the glass’. This is illustrated in Figure 5.22. SENTENCE CLAUSE

PARSER

CORE NP

NUC

NP

PRED V

1c1 Voice? -- Active ∴ PSA = Actor

Kim

smashed the glass.

Actor

Undergoer

3 2 LEXICON

Actor

Undergoer

[do¢ (x, Ø)] CAUSE [BECOME smashed¢ (y)]

Figure 5.22 Linking from syntax to semantics in simple English transitive clause

In step 1, step (c) is required only in languages with voice oppositions. In [S, AT ] languages this step would be unnecessary, and the linking from syntax to semantics in such languages is basically trivial: in a clause with a transitive verb, the privileged syntactic argument is always an actor and the other direct core argument is always the lower-ranking argument in terms of (4.14). The linking from syntax to semantics in Enga, a Papuan language of this type (Li and Lang 1979), is illustrated in Figure 5.23 for the example in (5.11). (5.11)

Baa-mé mená dóko-Ø p-´ı-á. 3sg-ERG pig DET-ABS kill-PAST-3sg ‘He killed the pig.’

The ergative case in Enga always marks the actor and the absolutive the undergoer. Hence the linking is simple and straightforward. The syntax-to-semantics linking algorithm for Enga would include only steps 1a–b, 2 (without a–b), 3 and 4 from (5.10); Enga has in situ WH-words, and therefore there is no need for step 5. This is the simplest possible syntax-to-semantics linking algorithm that a language could have, and that is perhaps one of the reasons why this language type is the most frequent cross-linguistically.

151

152


SENTENCE CLAUSE

PARSER

CORE NP

NP

NUC PRED V

1b

Baa-mé mená dóko p-í-á. Actor Undergoer

3 2 LEXICON

Actor

Undergoer

[do¢ (x, Ø)] CAUSE [BECOME dead¢ (y)]

Figure 5.23 Linking from syntax to semantics in Enga sentence in (5.11)

Steps 2a,b are required only in languages like English and Dyirbal in which variable linking to undergoer is possible with a single verb and there is no indication on the verb, e.g. an applicative morpheme, that a marked linking has occurred. Consider the following examples of the locative alternation in German. (5.12) a.

German laden/beladen ‘load’ Max hat die Kiste-n auf den Lastwagen geladen. have.3sgPRES the.FsgACC box-pl on the.MsgACC truck load.PSTP ‘Max [actor] loaded the boxes [undergoer] on the truck.’

a .

[do (Max, [load (Max, Kisten)])] CAUSE [BECOME be-on (Lastwagen, Kisten)]

b.

Max hat den Lastwagen mit den Kiste-n beladen. have.3sgPRES the.MsgACC truck with the.FsgDAT box-pl load.PSTP ‘Max [actor] loaded the truck [undergoer] with the boxes.’

b .

[do (Max, [load (Max, Kisten)])] CAUSE [BECOME be-on (Lastwagen, Kisten)], U=y

In the English translations of these two sentences, there are two different selections for undergoer, and there is no indication on the verb load as to whether an unmarked or a marked linking has occurred; the only clue is the preposition marking the oblique core argument. This is not the case in German, however: the verb laden ‘load [on]’ takes only the unmarked linking, i.e. the lowest ranking argument in the logical structure as undergoer, while the verb beladen ‘load [with]’ takes only the marked linking, i.e. the higher ranking non-actor argument. While these verbs would be related to each other in the lexicon via a lexical rule (cf. Van Valin and LaPolla (1997), section 4.6)), for linking purposes they count as distinct verbs, each of which takes only a single undergoer possibility. This would be indicated in the lexical entry for the verb lexicalizing the marked choice, i.e. ‘U = y’ for beladen ‘load [with]’. Hence steps 2a,b would not apply to them, unlike their English counterpart.


The sentence in Figure 5.12, Chris was presented with the flowers by Sandy at the party, exemplifies three additional complications: passive voice, variable undergoer selection, and an adjunct locative PP. The first step is to determine the voice of the verb; it is passive voice, which means that the privileged syntactic argument is not the actor. The NP after the verb is oblique, marked by with, and the final NP is marked by by, which following 1c1b2 in (5.10) makes it the actor. The next step is to call up the logical structure for present from the lexicon, [do (x, Ø)] CAUSE [BECOME have (y, z)]. Step 2 from (5.5) is then carried out, with the result that x = the actor. Because there is more than one possible undergoer choice with this verb, undergoer cannot be assigned, following 2 in (5.10). Rather, by step 2b in (5.10), the role of the oblique core argument the flowers is deduced from its prepositional marker; since with is not a locative preposition, the flowers must be linked to the second argument position in a twoplace state predicate. By step 3 the two actors may be linked; this leaves the nonactor MR in the clause and the y argument in the logical structure unassigned, and the two remaining arguments must be linked in order to satisfy the completeness constraint. The result is Sandy = actor = x, Chris = undergoer = y, and the flowers = z. There is additional unlinked material in the clause, namely the PP at the party. Step 4 in (5.10) comes into play here. The logical structure be-at (x, y) is retrieved from the lexicon; the logical structure of present serves as the second argument in the logical structure, and the NP the party functions as the first argument. This yields the logical structure be-at (party, [[do (Sandy, Ø)] CAUSE [BECOME have (Chris, flowers)]]). This completes the linking, and it may be represented graphically as in Figure 5.24; the circled numbers refer to the steps in (5.10). SENTENCE CLAUSE

PARSER

CORE CORE

CORE

> CLAUSE > CLAUSE SENTENCE

Figure 6.16 English core coordination

Figure 6.15 English core cosubordination

to be presented in section 7.3.1. The constituent and operator projections of these two sentences are not the same: the core nodes are dominated by a superordinate core node in cosubordination (Figure 6.15) but not in coordination (Figure 6.16). This reflects the fact that in core cosubordination a core operator such as modality must ‘obligation’ is shared by all cores, whereas it is not shared in core coordination. This is also illustrated in the Turkish examples in Figure 6.17 (= (6.34a)) and Figure 6.18 (= (6.34b)). The different scopes of the modal operators are explicit in these two representations. This follows from the point made at the beginning of this section, namely, that the defining feature of cosubordination in contrast to coordination is the shared operators at the level of juncture.

204

t he stru ctu re of com plex sen tences SENTENCE

SENTENCE

CLAUSE

CLAUSE

CORE

PRED

PRED

V

V

Gid

-ip

NP

PRO

NUC

NUC

V

NUC

NUC

CORE

CORE

V

V

PRO

V NUC CORE MOD

CLAUSE

MOD

TNS

CLAUSE

IF

IF

SENTENCE

Figure 6.17 Turkish core cosubordination

Figure 6.18 Turkish core coordination


kai fuone

PRED

CORE

SENTENCE

Fu

PRED

NUC

CLAUSE

NP

NUC

V

V

NP

NUC

Müzik dinle-yerek uyu- yabil- ir-im Ø.

gör-meli- yiz Ø.

CORE

CLM CORE

CORE

CORE CLM CORE


NUC < PERIPHERY NUC NUC NUC PRED PRED V V V kumeV

fie

va.

vazai

NUC

PER NUC

NUC

PRED

NUC PRED

PRED

V

V

V

V

ufu

furi

numu

akoe.

NUC

NUC

NUC
simultaneous events > sequential events > unspecified Causal hierarchy: physical > verbal > underspecified [non-defeasible] > inferred [defeasible] Participant’s mental disposition: intention > perception > belief > knowledge Necessarily shared participant hierarchy: Yes > No

In some sense the temporal hierarchy is the most fundamental: actions treated as phases of a single event are by definition going to be closer to each other semantically than actions treated as distinct events. In the causal hierarchy, ‘physical’ refers to some kind of physical action on the part of the causer on the causee,

211

212

t he stru ctu re of com plex sen tences

whereas ‘verbal’ means that the causer acts on the causee by means of speech; the other two involve no specification of the kind of causality, the difference between the two residing in whether the causation is explicit (underspecified) or not (inferred).9 The mental disposition hierarchy concerns what kind of cognitive state of a participant is involved, and finally the necessarily shared participant hierarchy says simply that two events which necessarily share a participant are more closely related than two which do not. The interaction of these hierarchies can be seen in the following examples.10 (6.47) a. b. c. d. e.

Harry shot Tom dead. Tom died, because Harry shot him. Having been shot by Harry, Tom died. After Harry shot him, Tom died. Harry shot Tom, and he died.

Only in (6.47a) are the shooting and Tom’s dying treated as phases of a single event; this is causative [1] in (6.38a). In some languages this may be expressed by a single verb, e.g. German erschiessen ‘shoot to death’, Lakhota wot’a ‘shoot to death’ [lit. ‘cause to die by action from a distance’]. In all of the other sentences the two are treated as distinct events, with explicit temporal sequences expressed in (c) and (d). The second example is explicit about the causal relationship, but it leaves the temporal parameters unspecified. In (6.47c–e) causality is nothing more than an inference; for example, all would be true of a situation in which Harry shot Tom, Tom survived the shooting but then developed a secondary infection while in the hospital and subsequently died. A case of underspecified causality would be Harry caused Tom to open the window; this is causative [2] in (6.38g). Here there is no indication whether the causation, which is explicit, is physical or verbal; it is compatible with both. What is often called ‘direct manipulative causation’ involves physical causality, actions which are phases of a single event and a necessarily shared participant. The hierarchy concerning a participant’s mental disposition, given in (6.45) is relevant to an additional important distinction in causality, namely the one illustrated in (6.48). (6.48) a. b.

9

10

Harry made Tom open the window. Harry had Tom open the window.

Causality is a huge topic, and ideas from a variety of approaches need to be integrated in a fully fleshed-out causal hierarchy, including Talmy (2000)’s theory of force dynamics. For a detailed study of the interaction between the syntactic and semantic hierarchies, on the one hand, and of the alteration between morphological and syntactic coding, on the other, see Guerrero Valenzuela (2004), an investigation of the syntax and semantics of complex sentences in Yaqui. These alternative expressions of causality illustrate that the higher a semantic relation is on the interclausal relations hierarchy, the more likely it is that there will be multiple ways to realize it syntactically. That is, the higher the tightest linkage relation realizing a particular semantic relation is, the more looser linkage relations will be available for alternative codings of it. In (6.47), (a) is an example of nuclear cosubordination, (c) is clausal cosubordination, (d) is clausal subordination, and (e) is clausal coordination.

6.7 Focus structure in complex sentences

An important aspect of the contrast in (6.48) concerns the mental disposition of the causee: whether he is acting according to his own intention, as in (b), or not, as in (a). This contrast is independent of the physical vs verbal opposition, as the following examples show. (6.49) a. b.

Harry ordered Tom to open the window. Harry asked Tom to open the window.

Both of these involve verbal causality, i.e. they are jussives, and they differ, as in (6.48), as to whether the causee is acting of his own accord or not. The psych-action and purposive relations in (6.37) both refer to a participant’s intentions; the former expresses a participant’s intention or other mental disposition toward an action involving him/her, while the latter conveys a participant’s intention to realize some state of affairs by means of an antecedent action. With psych-action there is a mental state and some kind of action, which may or may not temporally overlap. With purposive there are two actions which are necessarily sequential, and the participant’s intention relates to both of them; this is normally expressed constructionally, as there is no verb expressing the participant’s intention in the construction. Thus in characterizing these two interclausal semantic relations the mental disposition and temporal hierarchies play a role. The temporal hierarchy is crucial in distinguishing direct from indirect perception, which involves the second value on the mental disposition hierarchy. Direct perception entails that the perceptual event and the event perceived are simultaneous, while indirect perception requires that they not be. Hence in indirect perception the perceptual event necessarily follows the deduced event. The necessarily shared participant hierarchy is relevant, too: in direct perception there is a necessarily shared participant, as indicated in (6.38h), whereas there is none in indirect perception. Clausal junctures are the building blocks of texts and discourse, and the relations among clauses at the discourse level include those involving discrete events in the temporal hierarchy, as well as the kind of relations that figure prominently in discourse theories like Rhetorical Structure Theory (e.g. Matthiessen and Thompson 1988).11 The hierarchies in (6.43)–(6.46) constitute only an initial, tentative step toward decomposing the interclausal semantic relations hierarchy and providing an explanation for the upper two-thirds of it.12 The lower third overlaps with the semantic relations investigated in theories of discourse structure.

6.7

Focus structure in complex sentences

The focus structure projection of a sentence is related to the operator projection, because the nature of the speech act node which anchors the 11 12

See Winther-Nielsen (1995) for an analysis of Biblical Hebrew texts which combines the RRG theory of clause linkage with Rhetorical Structure Theory. See Guerrero Valenzuela (2004) for more discussion of these hierarchies and their interaction.

213

214


focus structure projection is expressed by the illocutionary force operator. This has critical implications for the potential focus domain in complex sentences. Illocutionary force cannot be independently specified in a subordinate clause of any kind; only clause nodes immediately dominated by the sentence node can have an illocutionary force operator. However, it is possible for some subordinate clauses to be within the scope of the illocutionary force operator, and therefore to be questioned or asserted (see Erteschik-Shir 1973; Erteschik-Shir and Lappin 1979), i.e. within the potential focus domain, and this brings up the important question of determining which types of subordinate clause can and cannot be within the potential focus domain. Both structural and lexical factors play central roles, and in this section only the structural considerations will be addressed; lexical factors will be discussed in section 7.6. There is a very general structural constraint governing the potential focus domain in complex sentences, and it is stated in (6.50). (6.50)

The potential focus domain extends into a subordinate clause if and only if the subordinate clause is a direct daughter of (a direct daughter of) the clause node which is modified by the illocutionary force operator.

In principle, there is no limit to the number of direct daughters involved, and therefore the specification in parenthesis should be considered to be recursive. This constraint can be illustrated in the contrast between Figures 6.12 and 6.6. In Figure 6.12, the subordinate clause is a direct daughter of the clause which is directly dominated by the sentence node; this is true in both the constituent and operator projections. Consequently, both clauses are in the potential focus domain in Figure 6.12, and this is represented in Figure 6.24. As in the figures in chapter 3, SENTENCE CLAUSE CORE NP

NUC

NP

PRED

CLAUSE

CLM

CORE

V Kim

told

Pat that NP

NUC

PERIPHERY PP ADV

PRED V she will arrive IU

IU

IU

IU

[IU

IU

at the party

late.

IU

IU]

SPEECH ACT Figure 6.24 Potential focus domain in clausal (daughter) subordination


the dotted line represents the potential focus domain, and ‘IU’ indicates the basic information units. A significant property of this construction is that there are two ways the embedded clause can be interpreted in information-structure terms. The subordinate clause as a whole may function as a single information unit, since it can be replaced by a WH-word, e.g. What did Kim tell Pat, or the elements inside of it may also be interpreted as distinct information units, because the potential focus domain extends into the subordinate clause. Evidence that the potential focus domain extends into it comes from possible answers to yes–no questions. In order to be the focus of a question (i.e. in the actual focus domain), an element must be in the potential focus domain, since only elements within the potential focus domain have the possibility of being in the actual focus domain. Consider the following question–answer pair.13 (6.51) Q: A:

Did Kim tell Pat that he will arrive at the party late? No, e arly.

It is possible for the focus of the yes–no question to fall on the subordinate clause peripheral adverb late, as the felicity of the response denying late and asserting early shows. Hence the subordinate clause must be in the potential focus domain, because in the question in (6.51) the actual focus domain is in the subordinate clause. In Figure 6.6, on the other hand, the subordinate clause bears no direct relationship to the matrix clause; it is the core argument of the predicative preposition after, and the PP is part of the peripheryCORE of the clause; in the operator projection, the clause is represented as an adjunct modifier of the core of the matrix clause. It is not a direct daughter of the matrix clause node. Hence the potential focus domain does not extend beyond the matrix clause (including the PP as a whole), and this is represented in Figure 6.25 on p. 216. The ad-core subordinate clause as a whole functions as a single information unit within the main clause and can be replaced by a WH-word, e.g. When did Pat see Kim?; its internal constituents are not, however, encompassed within the potential focus domain. This can be seen in the question–answer pairs in (6.52). (6.52) Q: A:

Did Pat see Kim, after she arrived at the party late? a. No, Sally. b. No, before. c. ??No, early.

The infelicity of the (c) response, in sharp contrast to the answer in (6.51), shows that the potential focus domain does not include the constituents of the ad-core subordinate clause. After can be the focus of a yes–no question, because it is part of the peripheral PP in the main clause. The principle in (6.50) correctly predicts 13

This test involves question–answer pairs in which the answer consists of ‘No’ plus a single constituent representing a single information unit. This single constituent is the focus in the answer and corresponds to the focus constituent in the question.

215

216



Pat

NUC

NP

PERIPHERY PP

PRED

NUC

CLAUSE

V

P

CORE

saw

Kim

NP

NUC

PP

PRED V after she IU

IU

IU

arrived at the party IU

SPEECH ACT Figure 6.25 Potential focus domain in ad-core subordination

that the that-clause in (6.51) is included within the potential focus domain and that the internal constituents of the ad-core clause in (6.52) are not. As with focus structure in simple sentences (see section 3.3), some languages overtly mark whether a clause in a complex sentence is within the potential focus domain or not, e.g. German (see Van Valin 1998). What kind of evidence can be used to show whether an embedded clause is contained within the potential focus domain or not, in a language without such overt marking? One example is whether a constituent can be the focus of a yes–no question as indicated by the felicity of a single constituent response, as in (6.51) and (6.52). Another type of evidence involves the distribution of focus-sensitive elements. For example, if a language has overt focus particles marking NPs in focus, then their distribution in embedded clauses would be a telling indicator of whether the potential focus domain extends into embedded clauses. If they can occur in a particular type of embedded clause with their focus-marking function, that is evidence that that clause type is within the potential focus domain. A clear example of this can be found in Lakhota. The relevant focus-sensitive element in Lakhota is a set of indefinite articles which are non-referential (non-specific); they can occur in only three contexts: in the scope of the question particle he, in the scope of negation, and in the scope of the ‘future tense’ morpheme -kte (Williamson 1984). The question operator will be used in this discussion; it occurs sentence-finally, as shown in (5.40) in section 5.4.3. This is illustrated in (6.53). (6.53) a.

14

Hokˇs´ıla eyá thaló ki ma-Ø-Ø-nú-pi.14 boy some [+SPEC] meat the INAN-3A-steal-pl

The Lakhota verbs manú ‘steal’, wa¸yá¸ke ‘see’, iyúkˇca¸ ‘think’ and ik´ıcˇ iˇcu ‘get for’ in (6.53)–(6.57) take their actor and undergoer markers as infixes.


a . *Hokˇs´ıla etá¸ thaló ki manúpi. some [–SPEC] ‘Some boys stole the meat.’ b. Hokˇs´ıla eyá thaló ki manúpi he? some [+SPEC] Q *‘Did some boys steal the meat?’, but OK: ‘Did some boys steal the meat?’ or ‘Did some boys steal the meat?’ b .

Hokˇs´ıla etá¸ thaló ki manúpi he? some [–SPEC] ‘Did some boys steal the meat?’, but *‘Did some boys steal the meat?’ or *‘Did some boys steal the meat?’

In the non-interrogative utterance in (a), only the indefinite–specific article eyá ‘some’ can occur; the non-specific indefinite article etá¸ is ungrammatical, as (a ) shows. In an interrogative utterance like (b), eyá is ungrammatical if its NP is the focus of the question; otherwise, it is grammatical. In (b ), on the other hand, etá¸ can only occur if its NP is the focus of the question; if the focus falls on a different constituent, the result is ungrammatical. The distribution of these two articles is evidence as to whether an NP can be in the actual focus domain in a particular clause, and this in turn is evidence as to whether the potential focus domain extends into the clause. In the following sentences, NPs with eyá and etá¸ occur in three types of subordinate clauses: object complement in (6.54), definite restrictive relative clause in (6.55),15 and ad-core subordinate clause in (6.56). Each sentence will have a question operator in the main clause, and if etá¸ can occur in the subordinate clause, then this shows that the focus of the yes–no question can be in the subordinate clause. If so, the internal constituents of the clause are in the potential focus domain, as in Figure 6.24. (6.54) a.

b. (6.55) a.

b.

15

[Hokˇs´ıla etá¸ thaló ki ma-Ø-Ø-nú-pi] i-Ø-Ø-yúkˇca¸ he? boy some [−SPEC] meat the INAN-3A-steal-pl INAN-3sgA-think Q ‘Does he think some boys stole the meat?’ *[Hokˇs´ıla etá¸ thaló ki manúpi] iyúkˇca¸. ‘He thinks some boys stole the meat.’ wiˇchá-Ø-yaxtake] ki le] *Wiˇchásˇa ki [[ˇsu´¸ka wa¸ igmú etá¸ man the [[dog a cat some [−SPEC] 3plU-3sgA-bite] the this] he? wa¸-Ø-Ø-yá¸ka 3sgU-3sgA-saw Q ‘Did the man see the dog which bit some cats?’ *Wiˇchásˇa ki [[ˇsu´¸ka wa¸ igmú etá¸ wiˇcháyaxtake] ki le] wa¸yá¸ke. ‘The man saw the dog which bit some cats.’

Lakhota relative clauses are internally headed, and the head inside the relative clause is obligatorily marked indefinite, with its actual definiteness value signalled by the article + demonstrative at the end of the relative clause.

217

218


c.

Wiˇchásˇa ki [[ˇsu´¸ka wa¸ igmú eyá wiˇcháyaxtake] ki le] wa¸yá¸ke. some [+SPEC] ‘The man saw the dog which bit some cats.’

d.

Wiˇchásˇa ki [[ˇsu´¸ka wa¸ igmú eyá wiˇcháyaxtake] ki le] wa¸yá¸ka he? ‘Did the man see the dog which bit some cats?’

(6.56) a.

*Wiˇchásˇa etá¸ Ø-wóta-pi eˇchúha¸, w´¸ı ya¸ ki mn´ı man some [−SPEC] 3A-eat-pl while woman the water i-Ø-w´ıcˇ ha-Ø-kiˇciˇcu he? INAN-3plU-3sgA-get.for Q ‘While some men were eating, did the woman get them water?’

b.

*Wiˇchásˇa etá¸ wótapi eˇchúha¸, w´¸ı ya¸ ki mn´ı iw´ıcˇ hakiˇciˇcu. ‘While some men were eating, the woman got them water.’

c.

Wiˇchásˇa eyá wótapi eˇchúha¸, w´¸ı ya¸ ki mn´ı iw´ıcˇ hakiˇciˇcu. some [+SPEC] ‘While some men were eating, the woman got them water.’

d.

Wiˇchásˇa eyá wótapi eˇchúha¸, w´¸ı ya¸ ki mn´ı iw´ıcˇ hakiˇciˇcu he? ‘While some men were eating, did the woman get them water?’

In (6.54a) etá¸ in the complement clause is grammatical, and this shows that the NP hokˇs´ıla etá¸ ‘some boys’ can be the focus of the yes–no question; hence, the internal constituents of the complement clause are within the potential focus domain. In the relative clauses in (6.55), however, etá¸ is impossible, regardless of whether the sentence is a question or not, and only eyá is grammatical. Consequently, it may be concluded that the internal constituents of the definite restrictive relative clause are not included in the potential focus domain. Finally, the same pattern is found in the ad-core subordinate clauses in (6.56) that is found in the relative clauses in (6.55): the occurrence of etá¸ in the embedded clause is ungrammatical and only eyá is possible, regardless of the illocutionary force of the sentence as a whole. Thus, the possibilities of occurrence of etá¸ and eyá in these embedded clauses lead to the conclusion that the internal constituents of object complements are within the potential focus domain but those of definite restrictive relative clauses and ad-core subordinate clauses are not. This conclusion is confirmed by looking at the range of possible felicitous answers to yes–no questions involving these same constructions. This is illustrated in (6.57). (6.57) a.

[Hokˇs´ıla etá¸ thaló ki manúpi] iyúkˇca¸ he? boy some [−SPEC] meat the steal think Q ‘Does he think some boys stole the meat?’ – Hiyá, wiˇch´¸ı cˇ ala eyá. no girl some [+SPEC] ‘No, some girls.’


b.

Wiˇchásˇa ki [[ˇsu´¸ka wa¸ igmú eyá wiˇcháyaxtake] ki le] wa¸yá¸ka he? man the dog a cat some [+SPEC] bite the this see Q ‘Did the man see the dog which bit some cats?’ – Hiyá, wa¸yá¸keˇsni. no, see.NEG ‘No, he didn’t see it.’ – Hiyá, w´¸ı ya¸ ki (wa¸yá¸ke). woman the (see) ‘No, the woman (saw it).’ – Hiyá, mathó wa¸ (wa¸yá¸ke). bear a (see) ‘No, (he saw) a bear.’ – *Hiyá, magá eyá (wiˇcháyaxtake). duck some (bite) *‘No, (it bit) some ducks.’

c.

[Wiˇchásˇa ki wóte] eˇchúha¸, tha-w´ıcˇ u ki mn´ı ik´ıcˇ iˇcu he? man the eat while his-wife the water get.for Q ‘While the man was eating, did his wife get him water?’ – Hiyá, Fred (mn´ı ik´ıcˇ iˇcu/*wóte). ‘No, Fred (got it for him)’, or ‘No, (she got it for) Fred’ *‘No, Fred was eating.’

In (a) the embedded clause is an object complement, and as the potentially appropriate response indicates, focus may fall within the embedded clause, since it is felicitous to deny the subject NP of the complement in the response. (Note the switch from etá¸ in the question to eyá in the answer.) This is further evidence in support of the conclusion reached above. In (b) the embedded clause is a definite restrictive relative clause. As always in this language, it is possible for the focus to fall on all of the matrix clause elements, including the NP interpreted as the head of the relative clause, but as the last response shows, it is impossible for an element within the relative clause to be the focus of the question. The final example involves an ad-core subordinate clause, and as the range of potential responses indicates, an element within the subordinate clause cannot be the focus of the question. Thus while the focus can fall on any major element of the matrix clause, it cannot fall within the embedded clause if it is a definite restrictive relative clause or ad-core subordinate clause, and this confirms the conclusion based on the distribution of the indefinite articles in these same constructions. The potential focus domain in these Lakhota constructions is summarized in (6.58). (6.58) a. b. c.

219

Summary of potential scope of he: potential focus domain [in boldface] [Hokˇs´ıla etá¸ thaló ki manupi] ´ iyukˇ ´ ca¸ he? Wiˇchásˇa ki [[ˇsu´¸ka wa¸ igmú eyá wiˇcháyaxtake] ki le] wa¸yá¸ka he? [Wiˇchásˇa ki wóte] eˇchuha ´ ¸, tha-w´ıcˇ u ki mn´ı ik´ıcˇ iˇcu he?

220

t he stru ctu re of com plex sen tences SENTENCE

SENTENCE CLAUSE CLAUSE

CLAUSE NP

CORE

CORE

NP CLAUSE

Hoksíla ˇ he ˇ etá thaló ki manúpi iyúkca Wichása wa igmú eyá wicháyaxtake ki le wayáka he ˇ ˇ ki súka ˇ ˇ CLAUSE (B) (A) CLAUSE IF CLAUSE IF

SENTENCE

SENTENCE

SENTENCE

CLAUSE PERIPHERY

CORE NP

PP CLAUSE Wich ˇ ása ˇ ki

wóte

NP

P ech ˇ ki mní ˇ úha, thawícu

CLAUSE

ikícicu he ˇ ˇ CORE

(C)

CLAUSE

IF

SENTENCE

Figure 6.26 The structure of Lakhota complex sentences in (6.58)

Does the principle in (6.50) predict this distribution of the potential focus domain in Lakhota? As can be seen from the simplified representations of these constructions in Figure 6.26, the answer is ‘yes’. The (A) diagram in this figure represents (6.58a), in which the embedded clause is a direct daughter of the clause node modified by the illocutionary force operator. As predicted, the potential focus domain extends into the embedded clause. The (B) diagram is the restrictive relative clause in (6.58b), and because the clause is embedded inside a NP, there is no relationship at all between the embedded clause and the clause node modified by the illocutionary force operator. Accordingly, the principle in (6.50) correctly predicts that the constituents of the embedded clause are outside of the potential focus domain. Finally the (C) diagram represents the ad-core subordinate clause in (6.58c), and as in the English ad-core subordinate clause discussed earlier, the embedded clause is an adjunct modifier of the core and not a direct daughter of the clause node modified by the illocutionary force operator. Hence, the principle in (6.50) correctly predicts that the internal constituents of the embedded clause are not within the potential focus domain. Thus, the distribution of eyá and etá¸ in Lakhota, the second of which must occur in the potential focus domain, is correctly predicted by the principle in (6.50).

6.8

Complex NPs

Relative clauses and noun complement constructions have not been discussed, because the theory of clause linkage applies to the joining of clauses and

6.8 Complex NPs

subclausal units with each other; these constructions involve the linking of a clause with a nominal head. However, the concepts from the theory of clause linkage can be extended to provide a framework for their analysis. This is summarized in Table 6.1. See Van Valin and LaPolla (1997: section 8.6) for detailed discussion. Table 6.1 Juncture and nexus in the NP Level of juncture

Nexus type(s)

Example

NP

Coordinate Cosubordinate Subordinate Cosubordinate Subordinate Cosubordinate Subordinate

the tall woman and the happy man the tall woman and happy man Chris, who loves soccer the order from the general to attack the city the rumour that Fred saw a UFO duckhunter, woodchopper the woman who left the party early

CoreN NuclearN

Noun complement constructions like the rumour that Fred saw a UFO involve a clause as a core argument of a nominal nucleus; this may be represented as in Figure 6.27. Hence it would be an example of coreN subordination. Restrictive relative clauses in a language like English, on the other hand, are non-argument, i.e. peripheral, modifiers of the nominal; it was argued in section 1.6 that restrictive modifiers occur at the nuclear level. Hence it is an example of nuclearN subordination. This is represented in Figure 6.28. NP COREN NUC N CLM CLAUSE N CORE the rumour NUC NP N PRED NUC N V NPROP that Fred

NP COREN NUC N N

saw

a

V

N

NUC

NUC N

CORE CLAUSE

UFO

TNS

COREN DEF NP

COREN DEF NP

Figure 6.27 Noun complement that-clause

Non-restrictive relative clauses have the same internal structure as restrictive relatives, but their structural relationship to the head noun is different. First, they are set off by a pause in languages like English, whereas restrictive relatives are

221

222

t he stru ctu re of com plex sen tences NP COREN NUC N

PERIPHERY CLAUSE

the two

N

PrCS

CORE

cars

NP

NUC

PERIPHERY

PROREL PRED

N

ADV

which AUX V

NUC N

were sold

NUM

COREN

yesterday

V NUC QNT

CORE

COREN

DEF

CLAUSE

NP

TNS

Figure 6.28 English restrictive relative clause

not. Second, as the name implies, they are not restrictive modifiers; their information content is additional information about the head noun, not information used by the interlocutor to determine the reference of the head noun. Third, since they often occur with proper nouns, which normally lack a layered structure, they cannot be in the periphery modifying the nominal nucleus, as in Figure 6.28. Hence they must be adjuncts at the NP rather than the nuclearN level. The structure of Chris, who loves soccer, is given in Figure 6.29. An important difference between restrictive and non-restrictive relative clauses is that non-restrictives PERIPHERY

NP

SENTENCE NPROP

CLAUSE CORE

PrCS

NP

NUC

NP PROREL

PRED

COREN

V

NUC N N

Chris,

who

loves

soccer

V

N

NUC

NUC N

CORE

COREN

CLAUSE TNS CLAUSE

NP

IF

SENTENCE

Figure 6.29 Structure of English non-restrictive relative clause

6.9 Syntactic templates: complex sentences

are embedded sentences, not clauses, because they have a distinct illocutionary force operator from the matrix clause. It is obligatorily declarative, since a nonrestrictive relative is an assertion about the head noun. Appositive XPs, e.g. Chris, a soccer lover, or Tom Smith, 81, are in the same relation to the head noun as non-restrictive relatives.

6.9

Syntactic templates for complex sentences

The notion of ‘syntactic template’ was introduced in the discussion of the syntax of simple sentences in chapter 1, and syntactic templates are also relevant to the analysis of complex sentences. Certain complex constructions can be handled in terms of the templates introduced in chapter 1, whereas others require new templates specific to them. The constructions for which no special templates would be required involve subordination, both daughter and peripheral. In core subordination, for example, a core or a clause functions as a core argument, and all that needs to be stated is that a node in a core can be filled by core or a clause, as in Figure 6.11. Nothing special needs to be specified for peripheral subordination involving ad-core and ad-clausal subordinate clauses, as in Figure 6.8. In the case of ad-core peripheral subordination, there is a predicative preposition in the peripheryCORE which takes a clause rather than a simple NP as its argument; the basic template for the matrix clause is the same regardless of whether the argument of the preposition is an NP or a clause. Ad-clausal subordinate clauses in the peripheryCLAUSE are marked by a clause-linkage marker rather than a preposition. Ad-nuclear modifying nuclei, as in Figure 6.9, involve a nucleus occurring in the peripheryNUCLEUS . In order to accommodate these constructions, it would be necessary to specify that clauses and nuclei, as well as PPs and adverbs, can occur as peripheral constituents. Clausal subordination involving ‘object complements’ (e.g. that-clauses), as in Figure 6.12, would require a specific template, which is given in Figure 6.30, because of the occurrence of a clause node as a direct daughter of a clause node. This template can iterate, with the whole template replacing the embedded clause node, producing a sequence of embedded complement clauses, as in Kim says that Chris thinks that Dana knows that Pat claims that Robin denies . . . CLAUSE CORE

CLAUSE

Figure 6.30 Template for clausal (daughter) subordination

There are general templates for coordination and cosubordination, independent of the level of juncture. They are given in Figure 6.31. The essential difference between the two templates is that in coordination the linked units are daughters of a node of the next highest layer of the clause, whereas in cosubordination they

223

224

t he stru ctu re of com plex sen tences UNITL i

UNITL i+1 UNIT

Li

UNIT

Li

Coordination

...

UNIT

Li

UNIT

Li

...

Cosubordination

Figure 6.31 Templates for non-subordinate nexus types

are daughters of a node of the same layer of the clause. This is true at all levels of juncture. Chomsky has often pointed out the human languages contain a potentially infinite set of sentences and achieve this with a finite set of mechanisms. In RRG, this aspect of linguistic creativity is captured by the theory of complex sentences: nuclei, cores, clauses and sentences can be combined recursively by means of the different nexus types to create the range of sentences that constitute a human language.

7

Linking syntax and semantics in complex sentences

7.0


In this chapter the linking between semantic representations and syntactic representations in complex sentences is explored. An important question to be investigated is the extent to which the linking algorithms proposed in chapter 5 for simple sentences must be modified to deal with complex sentences. Most complex sentences pose no particular difficulties for the linking system developed in chapter 5. Clausal junctures, for example, are composed of clauses, each of which links like an independent clause. Moreover, nuclear junctures act for linking purposes like simple clauses containing a complex predicate, and they basically follow the algorithms for simple sentences. The real challenge comes from non-subordinate core junctures with their obligatory sharing of a core argument, as discussed in section 6.2. Accordingly, linking in clausal and nuclear junctures will be examined first. After the discussion of linking in the different juncture–nexus types and in complex NPs, reflexivization in complex constructions will be investigated, and again the question arises as to the extent to which the principles proposed in section 5.3 will have to be modified to deal with these new phenomena. In the final section an account of the restrictions on so-called ‘long-distance dependencies’ involved in WH-question formation, topicalization and relativization will be proposed. These restrictions, which fall under the principle known as ‘subjacency’ in the generative literature, are significant for linguistic theory, for theories of language acquisition and for related theories of cognitive organization. 7.0.1

The semantics-to-syntax linking algorithm

The algorithm for linking semantic representations to syntactic representations for simple sentences which was developed in chapter 5 is repeated below. (7.1)

Linking algorithm: semantics → syntax 1. Construct the semantic representation of the sentence, based on the logical structure of the predicator. 2. Determine the actor and undergoer assignments, following the actor–undergoer hierarchy in Figure 4.4. 3. Determine the morphosyntactic coding of the arguments 225

226

linking s yntax and semantics: complex sentences

a. Select the privileged syntactic argument, based on the privileged syntactic argument selection hierarchy and principles in (4.14)–(4.15). b. Assign the arguments the appropriate case markers and/or adpositions. c. Assign the agreement marking to the main or auxiliary verb, as appropriate. 4. Select the syntactic template(s) for the sentence following the principles in (5.2). 5. Assign arguments to positions in the syntactic representation of the sentence. a. Assign the [−WH] argument(s) to the appropriate positions in the clause. b. If there is a [+WH] argument of a logical structure, 1. assign it to the normal position of a non-WH-argument with the same function, or 2. assign it to the precore or postcore slot, or 3. assign it to a position within the potential focus domain of the clause (default = the unmarked focus position). c. A non-WH argument may be assigned to the precore or postcore slot, subject to focus structure restrictions (optional). d. Assign the [−WH] argument(s) of logical structure(s) other than that of the predicator in the nucleus to 1. a periphery (default), or 2. the precore or postcore slot, or 3. the left- or right-detached position.

7.0.2

The syntax-to-semantics linking algorithm

The algorithm for linking syntactic representations to semantic representations for simple sentences which was developed in chapter 5 is repeated below. (7.2)

Linking algorithm: syntax → semantics 1. Determine the macrorole(s) and other core argument(s) in the clause. a. If the verb is intransitive, then assign the privileged syntactic argument either macrorole or direct core argument status, depending upon the language (language-specific) b. If the verb is transitive and the language lacks voice oppositions, determine the macroroles from case marking and/or word order (language-specific). c. If the language has a voice opposition, determine the voice of a transitive verb (language-specific): 1. If the construction is syntactically accusative: a. If it is the unmarked voice, the privileged syntactic argument is actor. b. If it is passive, the privileged syntactic argument is not the actor of the predicate in the nucleus; 1. the actor may appear as a direct core argument (language-specific); or

7.0 General considerations

2.

3. 4.

5.

2. the actor may appear in the peripheryCORE marked by an adposition or an oblique case (language-specific); or 3. if there is no actor in the core or the periphery, then replace the variable representing the highest ranking argument in the logical structure with ‘Ø’. 2. If the construction is syntactically ergative: a. If it is the unmarked voice, the privileged syntactic argument is undergoer. b. If it is antipassive, the privileged syntactic argument is actor; 1. the undergoer may appear as a direct core argument or as an oblique element (language-specific); 2. if there is no undergoer in the core or the peripheryCORE , then replace the variable representing the lowest ranking argument in the logical structure with ‘Ø’. 3. Assign macrorole status to the other direct core argument, if it is not dative or in an oblique case (language-specific). d. If the language is head-marking and there are independent NPs in the clause, associate each NP with a bound argument marker (language-specific). Retrieve from the lexicon the logical structure of the predicate in the nucleus of the clause and with respect to it execute step 2 from (7.1), subject to the following proviso: a. If the language allows variable undergoer selection and if there is more than one choice for undergoer, do not assign undergoer to an argument in the logical structure. b. Determine the linking of the non-macrorole core argument: 1. If there is a two-place state predicate in the logical structure and if the non-macrorole core argument is marked by a locative adposition or dative or a locative-type case, then link it with the first argument position in the state predicate in the logical structure and link the other non-actor core argument (if there is one) to the second argument position in the state predicate, or 2. If there is a two-place state predicate in the logical structure and if the non-macrorole core argument is not marked by a locative adposition or dative or a locative-type case, then link it with the second argument position in the state predicate and link the other non-actor core argument (if there is one) to the first argument position in the state predicate. 3. Otherwise, link the animate NP with the first argument position in the state predicate in the logical structure. Link the arguments determined in step 1 with the arguments determined in step 2 until all core arguments are linked. If there is a predicative adpositional adjunct, then retrieve its logical structure from the lexicon, insert the logical structure of the core as the second argument in the logical structure and the object of the adposition as the first argument. If there is an element in the pre- or postcore slot (language-specific),

227

228


a. Assign it the remaining unlinked argument position in the semantic representation of the sentence. b. And if there are no unlinked argument positions in the sentence, then treat the WH-word like a predicative preposition and follow the procedure in step 4, linking the WH-word to the first argument position in the logical structure.

7.1

Linking in clausal junctures

Since clausal junctures are made up of clauses, their linking properties are for the most part determined by the linking properties of the constituent clauses. Consider the following two examples of clausal juncture. (7.3) a. b.

Dana jogged through the park, and Kim waved to him. Bill went for a run, because he needed some fresh air.

In the first example, Dana jogged through the park and Kim waved to him are distinct clauses, and each is linked independently of the other, just as if each were a simple sentence on its own. The fact that there is a pronoun in the second clause referring (possibly) to Dana in the first clause does not affect the linking. The same is true in the (b) example, which involves peripheral subordination at the clause level; in it each clause links separately. A simplified logical structure for (b) is given in (7.4) (cf. (6.38o)). (7.4)

[Bill went for a run] because [he needed some fresh air]

There is nothing in the linking algorithm in (7.1) which correctly assigns because he needed some fresh air to the peripheryCLAUSE . This is not because this is a complex sentence; rather, it is because all of the adjuncts linked by the semanticsto-syntax algorithm in chapter 5 were ad-core adjunct PPs. There are ad-clausal PP adjuncts: if reason and concessive clauses occur in the peripheryCLAUSE , then the corresponding reason and concessive adjunct PPs, e.g. (6.25a,b), should be in it, too. Consequently, step 5d1 in (7.1) is inadequate, for both simple and complex sentences. It may be revised as follows. (7.5)

Revision of step 5d1 in (7.1) d. Assign the argument(s) or logical structure in logical structure(s) other than that of the predicator in the nucleus to 1. a periphery (default) a. If the representation is pred (NP/LS, LSMAIN ), where pred is a prepositional predicate, then assign the P + NP/Core/Clause to the peripheryCORE . b. If the representation is LSMAIN pred /connective NP/LS, then assign the P + NP or CLM + Clause to the peripheryCLAUSE .

7.1 Linking in clausal junctures

This says that if the semantic representation is like (6.38n), then the predicative preposition plus its object, be it an NP, a core or a clause, is an ad-core modifier and is assigned to the peripheryCORE . If, on the other hand, the semantic representation is like (6.38o–q) or the corresponding representation involving the predicative prepositions because of and despite, then the resulting adjunct PP or clause is linked to the peripheryCLAUSE . Step 4 in the syntax-to-semantics algorithm in (7.2) also needs revision, since it only deals with ad-core adjuncts. It can be reformulated as follows. (7.6)

Revision of step 4 in (7.2) 4. If there is an adjunct phrase in a periphery, a. and if it is in the peripheryCORE , then retrieve the logical structure of the predicative adposition from the lexicon, insert the logical structure of the core as the second argument in the logical structure and the object of the adposition as the first argument; b. and if it is in the peripheryCLAUSE , then link the adjunct PP or clause logical structure to the matrix logical structure via the semantic representation of the adposition or clause-linkage marker.

The rule in 4a yields a representation like (6.38n), while 4b results in a representation like those in (6.38o–q). An interesting issue is raised by sentences like the one in (7.7). (7.7)

Kimi worked on the assignment in the morning and afternoon.

i/*j

will finish it in the

Clausal junctures such as this are found in English topic chains like (4.17a) and in the Tepehua topic chain in (4.22). The traditional name for them in the syntax literature is ‘conjunction reduction’. The logical structure for (7.7) is given in (7.8). [be-in (morning, [(do (Kimi , [work.on (Kimi , assignment)]))])] & [be-in (afternoon, [[do (xi , Ø)] CAUSE [BECOME finished (3sgN)]])]

(7.8)

Only highly topical elements can receive zero coding, and therefore from the point of view of focus structure, constructions like (7.7) involve conjoined predicatefocus or narrow focus constructions in which the privileged syntactic arguments, which are pragmatic controllers and pivots, are topics (see (4.17)–(4.21)). The juncture–nexus type of this construction is clausal cosubordination, because illocutionary force must be shared across all conjuncts. This is illustrated in (7.9). (7.9) a. b.

*Did Kim work on the assignment in the morning, and will finish it in the afternoon? *Kim worked on the assignment in the morning, and will finish it in the afternoon?

In the first example, only the first conjunct is questioned, and the result is ungrammatical, while in the second, only the second conjunct is questioned, which is

229

230


likewise ungrammatical. The whole sentence must be interpreted as either an assertion or a question, and the most felicitous way to question it is Is it the case that Kim worked on the assignment in the morning and will finish it in the afternoon? The constituent structure representation of (7.7) is given in Figure 7.1. There is no core-initial NP position in the second clause, because the argument position in the logical structure containing what would be the privileged syntactic argument is occupied by a lexically unfilled variable obligatorily coindexed with the privileged syntactic argument of the previous logical structure. Because the referent of this variable is recoverable from the construction as a whole, the completeness constraint is not violated. SENTENCE CLAUSE CLM

CLAUSE CORE NP

NUC

PERIPHERY PP

Kim

CORE NUC

PRED V

CLAUSE PERIPHERY NP

PRED PP

worked on the assignment in the morning and

V will finish

PP it

in the afternoon.

Figure 7.1 Constituent projection for (7.7)

Since this is a clausal juncture, each of the clauses links separately, but the construction as a whole imposes a constraint on the linking in the non-initial conjuncts: the argument obligatorily controlled by the privileged syntactic argument of the previous clause must be selected as the privileged syntactic argument of its clause, even though it is not lexically instantiated (see section 4.3).1 This is captured in the constructional schema for this construction, which is given in Table 7.1 on p. 231. There are a number of new features in constructional schemas for complex sentences. First, there are specifications of the juncture type and the nexus type. Second, there is a specification of the construction type, with an abstract representation of its criterial features. In this case it is the occurrence of no privileged syntactic argument in the non-initial conjuncts and the obligatory coreference with the initial pragmatic controller. This requires that the semantic representation of each non-initial clause contains a lexically unfilled, obligatorily coindexed argument. ‘Unit template(s)’ refers to the template selection principles in (5.2) as well as any special properties of the syntactic templates of the constituent clauses, and special linking requirements are also specified. The schema in Table 7.1 may appear to be overly specific in many respects, but it is in fact very general and also represents the essential features of the analogous construction in Tepehua in (4.22). Comrie (1988) discusses this construction in 1

This construction is not a candidate for the discourse representation theory-based analysis proposed in section 5.4.1, because the antecedent of the pivot of the non-initial clause must be contained in the previous syntactically linked clause, regardless of the larger discourse context. Hence this antecedent (controller) is syntactically determined.


Table 7.1 Constructional schema for English ‘conjunction reduction’ Construction : English conjunction reduction Syntax: Juncture: clausal Nexus: cosubordination Construction type: conjunction [CL [CORE NPi [NUC . . .] . . .] . . .]1 , [CL [CORE—i [NUC . . .] . . .] . . .]2 , . . .CLM [CL [CORE—i [NUC . . .] . . .] . . .]n Unit template(s): (5.2) PSA: Clause 1: variable syntactic controller = pragmatic controller Clause 1 + n: variable syntactic pivot = pragmatic pivot Linking: controlled argument in clause 1 + n = pragmatic pivot (4.15a, c2) Morphology: CLM: coordinating conjunction or disjunction Semantics : Sequence of events sharing a common primary topical participant Pragmatics: Illocutionary force: shared across all conjuncts Focus structure: predicate focus in all conjuncts

Slavic languages like Croatian, which is a so-called ‘pro-drop language’ (see section 1.6), and shows that for coreference purposes it behaves like the English and Tepehua constructions. Hence it would appear that this schema would apply to these languages as well. Thus, the constructional schema in Table 7.1 can be taken to be the general schema underlying ‘topic chains’ in languages. In the construction in (7.7), the default focus structure is predicate focus, and the construction requires highly topical pragmatic privileged syntactic arguments. There is another clausal juncture construction in which the focus structure is just the opposite; that is, in this construction the privileged syntactic argument in the second clause must be focal and the remainder of the second clause must be non-focal. Traditionally called ‘VP ellipsis’, it is exemplified in (7.10). Kim is eating an ice cream cone, and Sandy is, too. (7.10) a. b. Sam washed his car, and Bill did, too. c. Leslie may go to the concert, and Pat may, too. d. Donna has been interviewed by Channel 7, and Pedro will be, too. d . *Donna has been interviewed by Channel 7, and Pedro will, too. e. Yolanda didn’t see the UFO, and Vanessa didn’t, either/*too. f. Yolanda didn’t see the UFO, but Vanessa did (*either/*too).

All of these sentences have focal stress on the NP in the second conjunct, which is the only lexical element in the constituent projection of the second conjunct. The auxiliary configurations need not be identical in the two clauses, as (d) shows, but when they are different, all of the diverging auxiliary elements must be present in the second conjunct, as (d ) shows. If the two conjuncts both contain negative

231

232


operators, as in (e), then either rather than too must be used. If the two clauses have different polarity, as in (f), then neither either nor too can be used. Not only may tense and other operators be different in the two clauses, but illocutionary force may also be different in each clause. This can be seen in (7.11). (7.11)

Mary is working on the assignment, but is Sam (*either/*too)?

When the illocutionary force is different across clauses, as when polarity is different, neither too nor either can occur. Accordingly, the nexus type is coordination. The syntactic structure of (7.10c) is given in Figure 7.2. Too is treated as a presuppositional adverb (Soames 1982), and consequently it must be a clausal modifier at the same level as the illocutionary force operator. The new feature of this representation is the constituent projection of the core of the second clause, since it contains only a single argument and the nucleus node dominates nothing and connects directly with its operator projection counterpart. Recall from chapters 1 and 6 that the constituent and operator projections are mirror images of each other, and accordingly there must be a full layered structure in each, due to the occurrence of a core argument in the constituent projection and at least two clausal operators (illocutionary force and tense) in the operator projection. SENTENCE CLM

CLAUSE CORE NP

NUC

CLAUSE PERIPHERY CORE

PP

NP

NUC

ADV

PRED V Leslie may go to the concert, and

Pat may,

too.

V NUC

NUC

CORE

CORE

STA CLAUSE CLAUSE TNS IF

CLAUSE

STA CLAUSE CLAUSE TNS IF

CLAUSE

SENTENCE

Figure 7.2 The syntactic representation of (7.10c)

The primary complexity in this construction lies in the linking between the syntactic and semantic representations. In the semantics-to-syntax linking, the elements in the semantic representation of the second clause which are identical to those in the semantic representation of the first clause are not mapped into the syntax. It might appear that this is technically a violation of the completeness constraint, since elements in the semantic representation of a clause


are not being mapped into the syntactic representation of that clause. It is not, however, as a close reading of the completeness constraint reveals. It is repeated in (7.12). (7.12)

Completeness constraint: All of the arguments explicitly specified in the semantic representation of a sentence must be realized syntactically in the sentence, and all of the referring expressions in the syntactic representation of a sentence must be linked to an argument position in a logical structure in the semantic representation of the sentence.

The crucial phrase is ‘must be realized syntactically in the sentence’; because the semantic information in the second clause is recoverable from the first clause, which is part of the same sentence, the constraint is satisfied.2 In the linking from syntax to semantics, the semantic representation of the second conjunct must be projected from the semantic representation of the first conjunct. The semantic representation of the first clause in (7.10a) is given in (7.13). (7.13)

IF DEC TNS PRES ASP PROG [do (Kim, [eat (Kim, ice cream cone)]) & INGR eaten (ice cream cone)]

This semantic representation is projected as the semantic representation for the second clause, with a variable replacing the argument corresponding to the privileged syntactic argument of the first clause; this yields (7.14). (7.14)

IF DEC TNS PRES ASP PROG [do (x, [eat (x, ice cream cone)]) & INGR eaten (ice cream cone)]

Since there is only one element in the constituent projection of the second clause, it will be linked to the x argument in (7.14), thereby giving the correct interpretation. When the operators in the two clauses are different, as in (7.10d) and (7.11), then only the logical structure(s) will be projected, as the operators are represented overtly in the clause. One final complication arises in examples like (7.10b) with respect to the interpretation of the pronoun his. It has long been noted that such sentences are ambiguous, because the second clause can be interpreted as meaning that Bill washed Sam’s car, that Bill washed his own car or that they washed some third party’s car, just as the full, unreduced sentence can. This follows from the fact that the pronoun his, unlike a reflexive, is not constrained to take a 2

There also seems to be discourse ‘VP’ ellipsis, as in (i). (i)

Speaker 1: Kim is eating an ice cream cone. Speaker 2: Sandy is, too.

In this instance the completeness constraint would be violated, since Speaker 2’s utterance is a different sentence from Speaker 1’s. While a complete solution to this problem is beyond the scope of this discussion, the approach proposed in section 5.4.1 using discourse representation theory would be able to handle the interpretation of the second utterance.

233

234


particular antecedent, and therefore can refer to any of the possible antecedents. It must be specified, then, that, when the semantic representation is projected from the first clause, no non-obligatory coindexing is projected. By ‘non-obligatory coindexing’ is meant coindexing not associated with reflexives, which require obligatory coindexing. Hence the semantic representation of the second clause of (7.10b) would be as in (7.15). (7.15)

IF DEC TNS PAST do (x, [wash (x, [have (3sgM, car)])]) & INGR --car)]) washed ([have (3sgM, ----

The 3sgM possessor can be construed as coreferential with x, with an NP in the previous clause, or with a discourse antecedent. In the following examples, the coindexing would be projected into the second clause and adjusted to reflect the fact that there is a different privileged syntactic argument in the second clause. (7.16) a. b.

Dana saw herself, and Sally did, too. Sam washed his own car, and Bill did, too.

Thus, if (7.17a) is the semantic representation for the first clause in (7.16a), then (7.17b) would be the projected semantic representation for the second clause. (7.17) a. b.

see (Danai , herselfi ) see (xj , x-selfj )

The constructional schema for English ‘VP’ ellipsis is given in Table 7.2 on the next page. These two constructions, ‘VP’ ellipsis and conjunction reduction, have been taken as evidence for the existence of a VP node in English clause structure, because the part of the clause left in the non-initial conjunct in conjunction approximates a VP and the part of the clause missing in the ellipsis construction also approximates a VP. It has, however, been possible to account for these constructions in terms of the interaction of the layered structure of the clause and focus structure, following the proposal sketched in section 3.5.

7.2

Linking in nuclear junctures

Many nuclear junctures have logical structures very much like lexical causative verbs and, not surprisingly, they have very similar linking properties. The argument structures of the component logical structures are pooled to yield the composite argument of the structure of the entire logical structure; hence their linking properties are the same as two- or three-argument lexical verbs. The linking in nuclear junctures with a linked intransitive verb is straightforward, whereas it becomes more complicated when the linked nucleus contains a transitive verb, as in (6.16b) from French, repeated below as (7.18a).

7.2 Linking in nuclear junctures

Table 7.2 Constructional schema for English ‘VP’ ellipsis Construction : English ‘VP’ ellipsis Syntax: Juncture: clausal Nexus: coordination Construction type: conjunction [CL [CORE . . . [NUC . . .] . . .] . . .] CLM [CL [CORE NP [NUC ]]] Unit template(s): clause 1 – (5.2) clause 2 – PSA only, remainder structurally empty Linking: voice must be the same in both conjuncts. Morphology: CLM: coordinating conjunction AUX in Clause 2: default ADV in clause 2, if same IF, polarity: too (positive polarity) / either (negative polarity) Semantics : Semantic representation of second clause is projected from the first clause: (1) The PSA argument is replaced by a variable; (2) All operators are projected, except those overtly present in Clause 2. (3) Obligatory coindexing is preserved. Pragmati cs: Illocutionary force: shared across all conjuncts (default) Focus structure: non-initial PSA must be focal, remainder of clause topical

(7.18) a.

b.

c.

Je fer-ai mang-er les gâteaux a` Jean. 1sgNOM make-1sgFUT eat-INF the.Mpl cakes DAT John ‘I will make John eat the cakes.’ Je fer-ai mang-er les gâteaux par Jean. 1sgNOM make-1sgFUT eat-INF the.Mpl cakes by John ‘I will have John eat the cakes.’ [do (1sg, Ø)] CAUSE [do (Jean, [eat (Jean, gâteaux)]) & INGR eaten (gâteaux)]

When the linked nucleus contains an intransitive verb, its single argument will be the undergoer. When it contains a transitive verb, on the other hand, the question arises as to which argument will function as undergoer. Since this is a nuclear juncture, the logical structure in (c) maps into a single core. By the actor–undergoer hierarchy, je ‘1sgNOM’ would be actor, and, of the remaining two arguments, the patient les gâteaux is clearly the lowest ranking argument with respect to the undergoer end of the hierarchy and therefore will be undergoer. That leaves the effector Jean as a non-macrorole direct core argument, and assuming that French has the same basic case-marking rules as other accusative languages, i.e. the ones in (4.25) and (4.29), it will be assigned dative case, which in French is realized

235

236


by the preposition a` . This accounts for (7.18a), and the linking from semantics to syntax is diagrammed in Figure 7.3. This basic linking accounts for what Comrie (1976) shows to be the predominant case-marking pattern found in this type of causative construction.3

Figure 7.3 Semantics to syntax linking in the French nuclear juncture causative in (7.18a)

There is a second linking possibility for the logical structure in (7.18c), and it yields (7.18b), in which Jean, the causee, is marked by par, the preposition which marks passive agents in French passive constructions. The difference between the (a) and (b) sentences in (7.18) is more than just the choice of preposition; as in a passive construction, the PP par Jean can be omitted, yielding Je ferai manger les gâteaux ‘I will have the cakes eaten’, or ‘I will have someone eat the cakes’, whereas the dative PP a` Jean cannot be omitted. Hence par Jean acts like an adjunct in the peripheryCORE , whereas a` Jean acts like a core argument. How can this be explained? Nuclear junctures have the logical structure of a transitive verb and therefore have actor and undergoer arguments. When the linked verb is intransitive, it contributes the argument that will function as undergoer. When it is transitive, it likewise contributes the argument that will function as undergoer, but it also contributes another argument which is a potential actor; indeed, if the embedded logical structure were to occur on its own, that argument would be the actor in the clause. What French allows, and what many other languages with this construction do not, is for the embedded transitive logical structure to 3

See Van Valin and LaPolla (1997: section 9.2.2), for detailed discussion of case marking in nuclear juncture causative constructions cross-linguistically. It should be noted that secondary object languages work differently with respect to undergoer selection in these constructions; see the discussion of (4.60) from Yaqui in section 4.5 above.

7.2 Linking in nuclear junctures

take two macroroles, just as if it were an independent logical structure. Since the effector of faire ‘make, cause’ is also an actor, a problem arises: how can there be two actors in a single core? The answer is, there cannot be, and accordingly the actor which does not serve as the privileged syntactic argument appears in the periphery marked by par, just as in a passive construction, in which the actor is not a core argument. Which actor functions as the privileged syntactic argument? The actor of the matrix logical structure, that of faire, appears as the privileged syntactic argument. This follows from the fact that this actor is the first effector in a causal sequence, and it was argued in section 4.5 that the first effector in a causal sequence has priority for actor; in the rare circumstance when there are two actors, as in (7.18b), this principle may be extended to give the actor of the first effector in a causal sequence priority for selection as the privileged syntactic argument as well, since, in syntactically accusative languages, the norm is for the highest ranking argument in the logical structure to function as the privileged syntactic argument. This linking is presented in Figure 7.4. It should be clear that the linking from syntax to semantics in both (7.18a,b) can be handled by the linking algorithm in (7.2) without modification. SYNTACTIC INVENTORY

4 SENTENCE CLAUSE CORE NUC

NP

NP

NUC

NUC

PRED

PRED

V Je

PERIPHERY

PP

V

manger les gâteaux par Jean.

ferai 5

PSA:NOM ACTIVE: 3sg 1 LEXICON

ACTIVE: par NP INF

Actor

Actor

3

ACC Undergoer

2 [do¢ (1sg, Ø)] CAUSE [do¢ (Jean, [eat¢ (Jean, gâteaux)]) & INGR eaten¢ (gâteaux)]

Figure 7.4 Semantics to syntax linking in the French nuclear juncture causative in (7.18b)

The constructional schema for the French nuclear juncture causative is given in Table 7.3 on the next page. In addition to the syntactic difference regarding the causee in these two constructions, there is a semantic contrast as well, one which is captured somewhat in the English translations by means of the contrast between ‘make’ and ‘have’ as the gloss for faire. Hyman and Zimmer (1976) argue that in the construction in (7.18a) the causee may be interpreted as not acting volitionally, whereas in the

237

238


Table 7.3 Constructional schema for French nuclear juncture causative construction in (7.18) Construction : French causative construction Syntax: Juncture: nuclear Nexus: cosubordination Construction type: serial verb [CL [CORE NP [NUC [NUC . . .] [NUC . . .]] NP . . .] . . .] Unit template(s): (5.2) PSA: none Linking: default; or if verb in NUC2 is transitive, then its logical structure may assign two MRs Morphology : none Semantics : [PREDNUC1 ] CAUSE [PREDNUC2 ] Prag m at i c s: Illocutionary force: unspecified Focus structure: unspecified

construction in (b) the causee may be interpreted as acting volitionally. Another way of putting this is that the secondary effector, the causee, can more easily be construed as an agent in (b) than in (a), in terms of the implicature theory of agency proposed in section 2.4.1. Thus, par encourages the agent implicature, while a` is basically neutral with respect to it. This difference in interpretation is evidence that the PP which is omitted in a sentence like Je ferai manger les gâteaux is par NP, not a` NP. The translation is ‘I will have the cakes eaten’ or ‘I will have someone eat the cakes’, not ‘I will make the cakes be eaten’ or ‘I will make someone eat the cakes’, which is the interpretation of causatives with the causee marked by a` . Resultative constructions like Vince has wiped the table clean in (6.30a) are the best examples of nuclear junctures in English. Like the French constructions discussed above, they have a causative meaning, and they in fact have a logical structure like that of a lexical causative accomplishment: activity causes change of state with result. The logical structure for this example is given in (7.19b). (7.19) a. b.

Vince has wiped the table clean. [do (Vince, [wipe (Vince, table)])] CAUSE [BECOME clean (table)]

Unlike the lexical causative accomplishments discussed in chapter 2, both the causing activity and the change of state and result are explicitly specified. The linking here is straightforward: Vince is the actor and the table the undergoer, and because English is a strict SVO language, the undergoer occurs between the two nuclei (see Figure 6.10). The constructional schema for this construction is given in Table 7.4.

7.3 Linking in core junctures

Table 7.4 Constructional schema for English resultative construction in (7.19a) Construction : English resultative construction Syntax: Juncture: nuclear Nexus: cosubordination Construction type: serial verb [CL [CORE NP [NUC [NUC1 . . .] NP [NUC2 . . .]]]], NUC2 = V Unit template(s): (5.2) PSA: none Linking: default Morphology: none Semantics : [PREDNUC1 ] CAUSE [PREDNUC2 ], PREDNUC2 [+static] Pragmat i c s: Illocutionary force: unspecified Focus structure: unspecified

This schema can be applied to Goldberg’s (1995) much-discussed example in (7.20). (7.20) a. b.

Chris sneezed the napkin off the table. [SEML do (Chris, [sneeze (Chris)])] CAUSE [BECOME NOT be-on (table, napkin)]

This is a resultative construction with two somewhat unusual features, compared with resultatives like (7.19): first, the first logical structure is a semelfactive rather than an activity logical structure, and second, the first verb in the construction is intransitive rather than transitive. The second nucleus contains a predicative PP (cf. Figure 1.6), and this is common in motion resultatives like Kim pushed the chair into the room. Despite these differences, the basic semantics of the constructions fits the constructional schema, and the linking is straightforward: Chris is the actor, the napkin is the undergoer, and off the table is the result PP. Note that there is no reason to claim that sneeze has suddenly become transitive in this construction; because the arguments of the component logical structures are pooled to create a composite argument structure in a nuclear juncture, the napkin is an argument of the whole logical structure but not one of sneeze. Thus the basic constructional schema for English resultative constructions handles sentences like (7.20) without modification.

7.3

Linking in core junctures

Neither clausal nor nuclear junctures have required any revision of the linking algorithms in chapter 5, but some types of core junctures do. There are two

239

240


basic types of core junctures, subordinate and non-subordinate, each with rather different linking properties. Subordinate core junctures in English are illustrated in (7.21); they both involve daughter subordination. (7.21) a. a . b. b .

Kim regretted Pam’s dying her hair green. IF DEC TNS PAST regret (Kim, [[do (Pam, [dye (Pam, her hair)])] CAUSE [BECOME green (her hair)]]) That Pam dyed her hair green shocked everyone. IF DEC TNS PAST [TNS PAST [do (Pam, [dye (Pam, her hair)])] CAUSE [BECOME green (her hair)]] CAUSE [feel (everyone, [shocked ])]

The embedded logical structure in (7.21a ) lacks clausal operators, and therefore it will be realized as a non-finite core, in this case a gerund. The one in (7.21b ), on the other hand, carries the clausal operator of tense, and therefore it will be instantiated as a tensed clause. The embedded logical structure links internally independently of the matrix logical structure, but as a whole unit it is part of the linking of the matrix logical structure, because the embedded logical structure is an argument of the matrix logical structure in the semantics and a core argument of the matrix predicate in the syntax. Thus, daughter subordination at the core level requires no revision of the linking algorithms from chapter 5. Peripheral subordination at the core level also does not require any revision of the linking algorithms in (7.1) and (7.2). (7.22) a. b.

Bill went for a run, after he finished his homework. be-after ([he finished his homework], [Bill went for a run])

In this example, Bill went for a run and he finished his homework are distinct clauses, and each is linked independently of the other, just as if each were a simple sentence on its own. The fact that there is a pronoun in the second clause referring (possibly) to Bill in the first clause does not affect the linking. A simplified logical structure for (b) is given in (7.22b). The first logical structure is the argument of after, while the second one is the logical structure of the matrix core, just as in a simple sentence like Sandy presented the flowers to Chris at the party in Figure 5.9. Step 5d1 in the semantics-to-syntax linking algorithm in (7.1) handles the assignment of after + clause to the peripheryCORE . The revised version in (7.5) accomplishes this as well. Non-subordinate core junctures do require revision of the linking algorithms. In chapter 6 it was argued that the criterial feature of non-subordinate core junctures is a shared semantic argument between or among the linked cores. This shared argument will require modification of the syntax-to-semantics linking algorithm. The discussion of non-subordinate core junctures begins with control constructions, and then moves on to matrix-coding (a.k.a. ‘raising to subject’, ‘raising to object’, ‘exceptional case marking’) constructions.


Control constructions

7.3.1

Examples of (obligatory) control constructions are given in (7.23). (7.23) a. b. c.

Chris tried to see Pat. Kim persuaded Pat to go to the party. Robin promised Sandy to wash the dishes.

As discussed in section 4.2, there is a syntactic argument missing from the linked core which must be interpreted as being the same as one of the syntactic arguments of the matrix core. The matrix core argument interpreted as being the same as the missing syntactic argument in the linked core is the controller. Example (7.23a) shows ‘subject’ control, since the controller is the ‘subject’ of the matrix core. The (b) sentence illustrates ‘object’ control, since the controller is the ‘object’ of the matrix core. Finally, the (c) example involves ‘subject’ control. Since ‘subject’ and ‘object’ have no theoretical status in this framework, it is necessary to find an alternative analysis using the appropriate theoretical terms, i.e. syntactic pivot or controller, actor or undergoer. The theory of obligatory control refers to hypotheses about how the controller of the missing syntactic argument in the linked unit is to be determined. One of the most striking facts about control phenomena is how consistent control properties are across languages, regardless of their typological differences. The following examples illustrate control phenomena from Dyirbal, a syntactically ergative language, and Lakhota, a head-marking language; the Dyirbal examples are from Dixon (1972, 1994). (7.24) a.

b.

(7.25) a.

b.

Bayi yaa-Ø walma-u wayd¸i-li. NM.ABS man-ABS get.up-TNS go.uphill-PURP ‘The man got up to go uphill.’

Dyirbal

ŋuma-ŋgu giga-n banagay-gu. Balan yabu-Ø baŋgul NM.ABS mother-ABS NM.ERG father-ERG tell-TNS return-PURP ‘Father told mother to return.’

Wówapi ki Ø-yawá i-bl-úthe. book the INAN-read stem-1sgA-try ‘I tried to read the book.’

Lakhota

Wówapi ki hená Ø-yawá-wiˇcha-wa-ˇsi. book the those INAN-read-3plU-1sgA-tell ‘I told them to read those books.’

Dyirbal does not express notions like ‘want’ or ‘try’ with complex sentence constructions, and accordingly a purpose construction is used to illustrate the first type of control construction. In Dyirbal and Lakhota, the control relations are analogous to those in the English examples in (7.23a,b), despite the major typological differences among the three languages. From a purely syntactic perspective, languages like Dyirbal present an interesting problem: verbs meaning ‘tell’ are supposed to be ‘object’ control verbs, yet

241

242


in Dyirbal, the absolutive NP, the ‘subject’ (privileged syntactic argument), is the controller. Dyirbal is not the only syntactically ergative language that shows this control pattern. It is also found in Sama, another syntactically ergative language; the examples are from C. Walton (pers. comm.); see also Walton (1986). (7.26) a.

Baya’ aku N-b’lli tinapay. want 1sgABS ANTI-buy bread ‘I want to buy some bread.’

b.

Logos ku iya N-k’llo daing ma si ina’. persuade 1sgERG 3sgABS ANTI-get fish OBL PM mother ‘I persuaded him to get fish for mother.’

c.

Janji’ ku iya bayad-an saung. promise 1sgERG 3sgABS pay-LOC tomorrow ‘I promised him to pay [him] tomorrow.’

Sama

The single NP in the matrix core in (a) is aku, the first singular absolutive pronominal form, and as the only NP it must be the controller of the missing argument in the linked core. The actor in the matrix core in the last two sentences is expressed by ku, which is the first singular, ergative pronominal form, and the undergoer in the matrix core in both is expressed by iya, the third singular absolutive pronominal form. In (b) the matrix verb is logos ‘persuade’, and accordingly there is expected to be ‘object’ control; but because Sama is syntactically ergative, the controller iya is in fact the syntactic ‘subject’ (privileged syntactic argument). In (c) the matrix verb is janji’ ‘promise’, and accordingly, the expectation is for ‘subject’ control. But again, because this is a syntactically ergative language, the controller is not in fact the subject; it is the non-‘subject’ actor, ku. In chapter 4 it was argued that Acehnese (Durie 1985, 1987) appears to lack grammatical relations (syntactic controllers and pivots) altogether. Yet the control facts in this language are just like those in all of the other languages looked at so far. (7.27) a.

b.

Geu-tém [(*geu-)taguen bu]. 3-want (3A-)cook rice ‘She wants to cook rice.’

Acehnese

Geu-yue lôn [(*lôn-)peugöt kuwéh]. A-order 1sg (1sgA-)make cake ‘(S)he ordered me to make a cake.’

In Acehnese, the control facts are exactly like those in English, Dyirbal, Lakhota and Sama. Thus, the basic control facts involving intransitive and transitive matrix verbs appear to be the same, regardless of whether the language is configurational like English or not, whether it has grammatical relations or not, or whether it is syntactically ergative or syntactically accusative. Accounting for the controller in sentences in which there is only one possible controller, such as all of the (a) examples, is trivial. The interesting problem is posed by the control facts in the (b) and (c) examples. All of these sentences


involving verbs like persuade and tell have undergoer control. That is, the undergoer of the matrix core is the controller of the missing syntactic argument in the linked core; it is the undergoer of the matrix core which is the shared argument with the linked core, the core argument which functions as a semantic argument in the logical structure of each core. In Foley and Van Valin (1984) it was argued that this follows directly from the semantics of the verbs involved, and in particular from the semantics of causation, which may be represented roughly as in (7.28). (7.28)

Actor acts on undergoer (by verbal or non-verbal means) → Undergoer does action

In the prototypical case, the actor acts on the undergoer by either verbal or nonverbal means with the intention that the undergoer do some action or be involved in some process or other change. Verbs denoting states of affairs in which the actor acts on the undergoer by non-verbal means are usually called ‘causative’ verbs, e.g. make, force or cause in English, whereas if the actor uses verbal means, the verbs denoting these states of affairs are called ‘jussive’ verbs, e.g. tell, order or persuade in English (cf. (6.37)). The resulting theory of obligatory control, proposed originally in Foley and Van Valin (1984), is stated in (7.29). (7.29)

Theory of obligatory control 1. Causative and jussive verbs have undergoer control. 2. All other (M-)transitive verbs have actor control.

This theory applies to matrix verbs which are (M-)transitive; if the matrix verb is (M-)intransitive, then the single argument will be the controller by default. This semantically based theory has numerous positive features. First, it applies without modification to all of the languages we have discussed, regardless of their typological characteristics. That is, because it is stated in terms of macroroles, it applies equally to Sama and to Lakhota, to Acehnese and Dyirbal and to English. Second, it predicts that sentences like (7.23c) should have actor control; that is, verbs like promise are not exceptional in this theory. In terms of (7.29), commissives are neither causative nor jussive and therefore should have actor control. Third, because the control facts are a function of the semantics of the matrix verb, they also carry over as preferences to other constructions involving these verbs, as illustrated in (7.30). (7.30) a. b.

Tom promised Sam that he would wash the car. Tom persuaded Sam that he should wash the car.

The pronoun he in the that-clause is technically free to refer to either of the NPs in the matrix core or even to a discourse referent, but the preferred interpretations of both of these sentences follow the predictions of the theory of obligatory control: Tom is the preferred antecedent for he in (a), and Sam is the preferred antecedent for he in (b). Fourth, because the choice of the controller is tied to the semantics of

243

244


the verb, this analysis predicts that if a verb can be used alternatively as causative or non-causative or as jussive or non-jussive, then its control properties should change. This seems to be the case with a number of verbs of saying. (7.31) a. b. c.

Larry asked Sally to leave. Larry made a polite request to Sally that she leave. Larry requested permission from Sally so that he could leave.

For many English speakers (7.31a) is ambiguous, and the two interpretations are given in (b) and (c). Ask can be construed either as a jussive verb, as in (b), or a verb for requesting something, as in (c); when it has a jussive interpretation, there is undergoer control, as in (b), and when it is non-jussive, there is actor control, as in (c), exactly as predicted. Fifth, because control choices are characterized in terms of macroroles, rather than grammatical relations, the behaviour of these verbs under passivization is predicted. Consider the examples in (7.32). (7.32) a. b.

Pat was persuaded (by Kim) to go to the party. *Sandy was promised (by Robin) to help with the party.

(cf. (7.23b)) (cf. (7.23c))

Recall that the controller is the syntactic argument in the matrix core which also functions as a semantic argument in the linked core, and this entails that the controller must be a core argument in the matrix core. With a jussive verb like tell or persuade, the undergoer functions as privileged syntactic argument in a passive construction, and since it is a core argument, it continues to function as controller and the resulting sentence is fine, as (a) shows. As Cutrer (1987, 1993) points out, however, with a non-jussive verb like promise, the actor functions as an oblique peripheral constituent, not a core argument, in a passive construction, and consequently there is no core-argument controller in the matrix core; the resulting sentence is therefore ungrammatical, as (b) shows. The fact that ‘subject’-controlled complements cannot be passivized is known in the literature as ‘Visser’s generalization’, and it follows directly from the account of passive in chapter 4 and the theory of obligatory control in (7.29). The theory of obligatory control in (7.29) and the completeness constraint in (7.12) play crucial roles in both phases of linking in these constructions. On the syntactic side, the linked core in a non-subordinate core juncture is missing a syntactic argument position, which is the syntactic pivot of the construction. In order to capture the fact that there is a syntactic argument slot missing in the linked core, it is necessary to add a universally valid qualification to the syntactic template selection principle in (5.2), namely, the occurrence of the core as the linked core in a non-subordinate core juncture reduces the number of core slots by 1. The revised principles are given in (7.33). (7.33) a.

Syntactic template selection principle (revised formulation): The number of syntactic slots for arguments and argument-adjuncts within the core is equal to the number of distinct specified argument positions in the semantic representation of the core.


b.

c.

Universal qualification of the principle in (a): The occurrence of a core as the linked core in a non-subordinate core juncture reduces the number of core slots by 1. Language-specific qualifications of the principle in (a): 1. All cores in the language have a minimum syntactic valence of 1. 2. Argument-modulation voice constructions reduce the number of core slots by 1. 3. The occurrence of a syntactic argument in the pre/postcore slot reduces the number of core slots by 1 (may overrride c1).

The statement in (b) does not specify which syntactic slot is missing, since that is a construction-specific feature. The linked core in the constructions in (7.23) will always be ‘subjectless’, and this is the central fact about these constructions that the linking system must accommodate. The discussion of linking in control constructions will begin with the simplest example, (7.23a); its logical structure is given in (7.34b). (7.34) a. b.

do (xi , [try (xi , [see (yi , z)])]) do (Chrisi , [try (Chrisi , [see (yi , Pat)])])

The verb try takes a non-subordinate core juncture because it is a psych-action verb, in terms of the interclausal relations hierarchy in Figure 6.22; this follows from the meaning of psych-action: the verb codes a mental disposition on the part of its actor to be involved in a state of affairs, and accordingly the actor must also be a semantic argument of the embedded logical structure, because the participant with the mental disposition must also be a participant in the state of affairs denoted by the embedded logical structure. Thus the actor of try must control the missing argument in the linked core, i.e. it must also function as a semantic argument in the embedded logical structure. Hence Chris, the only argument in the matrix core, is the controller. The crucial syntactic feature of non-subordinate core junctures is that the linked core lacks a syntactic argument position, following (7.33b). The missing argument is the syntactic pivot of the linked core, and in these constructions the pivot is the traditional ‘subject’; therefore it is the prenuclear core argument position that is missing in the linked core. Because there is an obligatorily shared semantic argument in the construction, one of the arguments in the embedded logical structure is not filled by lexical material but is coindexed with the controller in the matrix logical structure. The result is the logical structure in (7.34b), and how it links into the syntax is presented in Figure 7.5 on the next page. Given the logical structure in (7.34b), the linking algorithm in (7.1) can handle this case without modification. Even though the y argument in the embedded logical structure is not directly linked to an expression in the syntax, it is coindexed with Chris, which is linked to the syntax, thereby satisfying the completeness constraint. There are other coindexing possibilities in the logical structure in (7.34); they are given in (7.35).

245

246


SYNTACTIC INVENTORY

4 SENTENCE CLAUSE CORE

NP

Chris

CORE CLM

CORE

NUC PRED

NUC PRED

V

V

tried

to

see

NP

Pat.

5 PSA:NOM ACTIVE: 3sg

1 LEXICON

Actor

ACTIVE: INF

ACC

3

Undergoer 2

do¢ (Chrisi, [try¢ (Chrisi, [see¢ (yi , Pat)])])

Figure 7.5 Linking from semantics to syntax in English control construction in (7.23a)

(7.35) a. *Chrisi tried [Pat] to see i . a . do (Chrisi , [try (Chrisi , [see (Pat, zi )])]) b. Chris tried to be seen by Pat. b . do (Chrisi , [try (Chrisi , [see (Pat, zi )])]) c. *Chrisi tried [Pat] to be seen [by i ]. c . do (Chrisi , [try (Chrisi , [see (yi , Pat)])])

For each of the possible coindexings, there is only one grammatical output. What rules out the impossible linkings? The answer is, the completeness constraint. In (a), the actor of try is coindexed with the second argument of see. This means that Chris is the actor of try and Pat is the actor of see, and when these assignments are mapped into the syntactic structure in Figure 7.5 a problem immediately arises: there is no syntactic position in the second core for an actor to appear in, since the second core has an active voice verb. Hence Pat cannot be realized in the syntactic representation, and this violates the completeness constraint. This is given in Figure 7.6. It is impossible to link an actor to the post-nuclear core argument slot, since actors may only be linked to the generalized privileged syntactic argument (‘subject’) position in an active-voice core in English. Hence no special constraints or principles are needed to explain the impossibility of this potential linking. The other two sentences involve passive voice in the second core, and here too only one of the two possible linkings is permitted. The sentence in (7.35b) has the same logical structure as in (7.35a), but the second core is passive. This


4

SYNTACTIC INVENTORY

SENTENCE CLAUSE CORE CORE CLM

CORE

NUC PRED

NUC PRED

V

V

NP

Chris

tried

to

NP

see ???


INF

Actor

1 LEXICON

ACTIVE:

ACC

3

Actor 2

do¢ (Chrisi, [try¢ (Chrisi, [see¢ (Pat, zi)])])

Figure 7.6 Failed linking from semantics to syntax in (7.35a)

makes possible a linking which does not violate the completeness constraint, as Figure 7.7 shows. The final possible combination is the syntactic representation of Figure 7.7 with the logical structure in (7.34); this combination cannot be successfully linked, as Figure 7.8 shows. The first thing to note here is that the sentence Chris tried to be seen is perfectly grammatical, but it is not a

SYNTACTIC INVENTORY

4 SENTENCE CLAUSE CORE CORE CLM NP

CORE PERIPHERY

NUC

NUC

PRED

PRED

AUX

V Chris

tried

PP

V to be

seen

by Pat.

5 PSA:NOM

1 LEXICON

ACTIVE: 3sg

by:ACC

PASS: Actor

3

INF

Actor

do¢ (Chrisi, [try¢ (Chrisi, [see¢ (Pat, zi)])])

Figure 7.7 Linking from semantics to syntax in (7.35b)

2

247

248


SYNTACTIC INVENTORY


NP

CORE CLM

CORE

NUC

NUC

PRED

AUX

PRED

V Chris

tried

V to

be

seen ???

PASS:

ACC


1 LEXICON

Actor

INF

3

Undergoer 2

do¢ (Chrisi, [try¢ (Chrisi, [see¢ (yi , Pat)])])

Figure 7.8 Failed linking from semantics to syntax in (7.35c)

possible realization of the logical structure in (7.35c ). Its logical structure is do (Chrisi , [try (Chrisi , [see (Ø, zi )])]), where the actor of the embedded logical structure is unspecified. The linking in Figure 7.8 violates the completeness constraint, because the undergoer Pat cannot be realized overtly in this syntactic structure. An undergoer in English can only be realized as a core argument, either in its default post-nuclear position or as privileged syntactic argument in a passive, and neither of these options is available in this structure. The only possible realization for this logical structure is as in Figure 7.5, as noted above. The four possibilities in (7.34) and (7.35) have been accounted for in terms of the linking algorithm in (7.1) and the completeness constraint; no special principles or constraints are required. In particular, it is not necessary to stipulate in the lexical entry for try that it must share an argument with its complement logical structure. Thus, there is no need to rule out via stipulation that do (Chris, [try (Chris, [see (Dana, Pat)])]) is an impossible logical structure, since there is no possible linking between it and any of the syntactic structures in Figures 7.5– 7.8 that could satisfy the completeness constraint. Moreover, the same is true with respect to the more complex examples in (7.23b,c); they are handled in exactly the same way. The only complexity they raise is that the theory of control in (7.29) specifies which argument in the matrix logical structure is coindexed with the lexically unfilled argument in the embedded logical structure. No stipulations would be required regarding which argument in the embedded logical structure would be lexically unfilled and coindexed with the argument in the matrix logical


structure, because the correct coindexings fall out exactly as in (7.34) and (7.35) (as in Figures 7.5–7.8). The problem of linking from syntax to semantics in these constructions requires the theory of obligatory control, because the correct interpretation of the sentence depends upon the correct assignment of the controller in the matrix core. Since the issue of the controller is trivial in clauses with M-intransitive matrix verbs, attention will be focussed on clauses with M-transitive matrix verbs like promise and persuade in English and other languages. The syntax to semantic linking in the Dyirbal example in (7.24b), following the algorithm in (7.2), is presented in Figure 7.9; the numbers refer to the steps in (7.2). The bulk of the linking is just as for the simple sentences discussed in chapter 5. Having determined the voice of the verb, the NPs functioning as actor and undergoer can be identified (step 1c2), and after accessing the lexical entries for the verbs in the lexicon and constructing the composite logical structure for the whole sentence, the actor and undergoer assignments for the arguments in it can be determined (step 2). The third step is to match the actor and undergoer of gigal ‘tell’ in the syntax with the actor and undergoer of its logical structure. At this point, all of the relevant steps in (7.2) have been executed, and yet there is, crucially, an unlinked argument, the actor of banagay ‘return’; the completeness constraint remains unsatisfied. This is where the theory of obligatory control in (7.29) comes into play: because gigan ‘tell’ is a jussive verb, its undergoer is the controller, and accordingly, the undergoer of gigan is linked to the actor of banagay, yielding the correct interpretation of the sentence and satisfying the completeness constraint. The obligatory control linking is represented by the thick, solid black line.

SENTENCE

PARSER

CLAUSE

NP

NP

1c2 Balan yabu baŋgul ŋumaŋgu Voice? −Active ∴ PSA = Undergoer 1b Undergoer Actor

CORE NUC

CORE

PRED

PRED

V

V

gigan

banagaygu

NUC

Theory of obligatory control

3 Actor

Undergoer

Actor

2 LEXICON

[do¢ (x, [say¢ (x, y)])] CAUSE [do¢ (z, [return¢ (z)])] Figure 7.9 Linking from syntax to semantics in Dyirbal control construction in (7.24b)

249

250


It is clear that it is necessary to add a step to the syntax to semantics linking algorithm to accommodate the crucial role that the theory of obligatory control plays. Accordingly, the following step may be added to (7.2); it follows step 3, and consequently step 4 in (7.2) is now step 5 and step 5 in (7.2) is now step 6. The final version of the linking algorithm will be given in (7.74). (7.36)

4. In non-subordinate core junctures, one of the arguments of the matrix core must be linked to an argument position in the embedded logical structure, following (7.29).

The constructional schema for the English obligatory control construction is given in Table 7.5. The schema covers all of the constructions with persuadeand promise-type verbs discussed in this section. The controller is in core 1 and is determined by (7.29). The pivot of the construction is in core 2, not core 1, and it is the highest ranking core macrorole. The linking is specified as ‘default’, as it follows (7.36) without modification. Table 7.5 Constructional schema for English control constructions in (7.23) Construction : English obligatory control construction Syntax: Juncture: core Nexus: co(sub)ordination Construction type: serial verb [CL [CORE NP [NUC . . .] (NP)] CLM [CORE [NUC . . .] . . .] . . .] Unit template(s): Core 1: (7.33a,c) Core 2: (7.33a,b,c) PSA: Core 1: controller = semantic controller, following (7.29) Core 2: pivot = variable syntactic pivot (4.15a,c2) Linking: Default Morphology : CLM to, from or Ø Semantics : psych-action, causative/jussive; commissive, directive speech acts Prag m at i c s: Illocutionary force: unspecified Focus structure: unspecified

7.3.2

Matrix-coding constructions

Matrix-coding constructions have gone by a number of names in the history of linguistics; the term ‘matrix coding’ is taken from Frajzynger (1995), who proposed it as a theory-neutral label. There are two basic types of matrixcoding constructions: what is called ‘raising to subject’ in the generative literature, as in (7.37a), and the construction in (b) which was originally called ‘raising to


251

object’ in transformational grammar but is known in the principles and parameters literature as the ‘exceptional case marking’ construction. (7.37) a. a . b. b .

Mary seems to know the answer. It seems that Mary knows the answer. Pat believes Chris to have eaten the bagel. Pat believes that Chris ate the bagel.

Each of these constructions has an alternative form in which there is a finite that-clause complement, and in both the core argument which is the ‘subject’ of the finite embedded clause in the alternative construction appears as a core argument in the matrix core, as ‘subject’ in (7.37a) or ‘object’ in (b), hence the names from transformational grammar. The two sentences in each pair have the same logical structure but different semantic representations. This is illustrated for (7.37b,b ) in (7.38), in which the logical structure for both sentences is given in (a) and abbreviated semantic representations for them are given in (b) and (c). (7.38) a. b. c.

believe (Pat, [do (Chris, [eat (Chris, bagel)]) & INGR eaten (bagel)]) IF DEC TNS PRES believe (Pat, [ASP PERF [do (Chris, [eat (Chris, bagel)]) & INGR eaten (bagel)] IF DEC TNS PRES believe (Pat, [TNS PAST [do (Chris, [eat (Chris, bagel)]) & INGR eaten (bagel)]

= (7.37b) = (7.37b )

The lack of the obligatory tense operator modifying the embedded logical structure in (b) entails that it will not be realized as a tensed clause, i.e. as a that-clause. Hence the semantic representation in (b) is for a core juncture. In (c), on the other hand, there is a tense operator modifying the embedded logical structure, and therefore it will be realized as a tensed clause. There are subtle differences in meaning between the two forms, which have been investigated in, e.g., Borkin (1984) and Langacker (1995). The ‘raising to subject’ or ‘matrix-coding as privileged syntactic argument’ construction is illustrated in (7.39) from Icelandic (Thráinsson 1979) and Kinyarwanda (Bantu; Kimenyi 1980); cf. also the examples from Acehnese in (4.5). (7.39) a.

b.

b .

Harald-ur virð ist haf-a far-ið heim. Harold-MsgNOM seem.3sgPRES have-INF go-PSTP home ‘Harold seems to have gone home.’

Icelandic

Abá-nyéshuur ba-kwii-ye gu-some ibitabo. Kinyarwanda 2-student 2-essential-ASP INF-read 8-book ‘Students must read the books.’ Bi-rá-kwii-ye ko abá-nyéeshuûri ba-sóm-a ibitabo. 8-PRES-essential-ASP CLM 2-student 2-read-ASP 8-book ‘It is essential that students read the books.’

252


There is no Icelandic counterpart to (7.37a ) involving the verb virðast ‘seem’ plus a tensed complement. The counterpart to the Kinyarwanda matrix-coding construction is given in (b ). The primary predicates which allow this construction in English are seem, appear, be likely and be certain, while only virðast ‘seem’ allows it in Icelandic, according to Thráinsson (1979). Kimenyi (1980) describes the verbs that occur in this construction in Kinyarwanda as ‘modality impersonal verbs’ (e.g. -shobok- ‘be possible’, -kw´ıi- ‘be essential’ and -bujijw- ‘be forbidden’), and factitive verbs like -babaj- ‘be sad’ and -taangaj- ‘be fascinating’. The essential feature of the verbs in this construction is that they are either atransitive, like English seem and Icelandic virðast, or they are intransitive like English be likely and be certain. The logical structure for English seem and Icelandic virðast is seem (x, y) [MR0], where the x argument is a perceiver which is optionally realized in English by a to PP and in Icelandic by a dative NP, as illustrated by the Icelandic example in (7.40) and its English translation. (7.40)

Harald-ur virð ist mér ver-a besti dreng-ur. Harold-MsgNOM seem.3sgPRES 1sgDAT be-INF best boy-MsgNOM ‘Harold seems to me to be a nice guy.’

The y argument is a proposition; hence it is filled by another logical structure. The occurrence of an argument from the embedded logical structure in the core headed by seem comes about as follows. There is a direct core argument slot in the matrix core, but the verb has no arguments which can fill it; moreover, since this is a core juncture, the second core is missing the prenuclear core argument position, following (7.33b). The semantic argument that would normally function as the privileged syntactic argument in the second core cannot be realized in it, due to the absence of its syntactic slot, and this would normally lead to a completeness constraint violation, as in, e.g., Figure 7.6. However, there is an open core argument position within the clause, namely the one in the matrix core. The argument may be linked to this position, thereby avoiding a completeness constraint violation. This linking across a core boundary is permitted by step 5a in (7.1), because it specifies the assignment of [−WH] arguments ‘to the appropriate positions in the clause’. This step in the linking is clause-bound but not core-bound; hence this linking is allowed. The linking from semantics to syntax in (7.37a) is given in Figure 7.10 on the next page. Mary is the actor of know but a core argument in the core headed by seem. The syntactic structure in Figure 7.10 is similar to that in the control constructions with try in Figure 7.5, but differs in terms of nexus. Try-constructions are cosubordinate, due to the shared deontic modality operator across the two cores. Sharing a core operator across the two cores is ruled out in principle in this construction, however, because the matrix predicate does not have an argument that can be modified by a deontic modal operator. Hence the nexus is coordinate. The essential difference between the two constructions lies not in the syntactic structure but rather in how the linking works, which is primarily a function of the semantic properties of the predicate in the matrix core.


4

SYNTACTIC INVENTORY

SENTENCE CLAUSE CORE CLM NP

Mary

CORE

NUC

NUC

PRED

PRED

V

V

seems

to

know

NP

the answer.

5 1 LEXICON

ACTIVE: 3sg

ACTIVE: PSA:NOM INF

ACC

Actor

Undergoer

seem¢ (Ø, [know¢ (Mary,

answer)])

3 2

Figure 7.10 Linking from semantics to syntax in English matrix-coding construction in (7.37a)

The linking from syntax to semantics in this construction requires only a minor modification of the linking algorithm in (7.2) and (7.36). The linking for What does Mary seem to like? is given in Figure 7.11. As before, the numbers refer to the steps in the linking algorithm in (7.2) as modified in (7.36). Step 1 applies in each core; since the first core is intransitive, the only conclusion that can be drawn is that Mary is a macrorole argument; there is no evidence in the first core SENTENCE

PARSER

CLAUSE PrCS

NP

NP

What does Mary

CORE CLM

CORE

NUC

NUC

PRED

PRED

V

V

seem

to

like?

6 MR

1c1 Voice?−Active ∴ PSA = Actor

4 Actor Undergoer

LEXICON

seem¢ (Ø, [like¢ (x, y)])

2

Figure 7.11 Linking from syntax to semantics in English matrix-coding construction

253

254


as to whether it is actor or undergoer. In the infinitival core the voice is active, and therefore the privileged syntactic argument NP would be the actor, if there were one in the core. Step 2 is straightforward. In order to execute step 3, the information from step 1 must be used. The core argument Mary cannot be linked to an argument position in the logical structure of seem, because the first argument is unspecified and the second one is filled by a propositional logical structure. There are no core arguments in the second core to be linked to the embedded logical structure like (x, y), and therefore Mary can be linked to an argument position in the embedded logical structure; the question is, which one? As in the control constructions discussed above, the obligatorily missing argument in the linked core is the syntactic pivot, and since step 1 revealed that the privileged syntactic argument of the second core is the actor, Mary must be linked to the actor argument in the semantic representation. Finally, the WH-word in the precore slot is linked to the remaining unlinked argument position, following step 6. The result is the correct linking, with Mary interpreted as the actor of like and what as the undergoer. The only modification of (7.36) that is required is to specify that, in step 4, the theory of obligatory control in (7.29) applies only to control verbs. This may be stated as in (7.41). (7.41)

4. In non-subordinate core junctures, one of the arguments of the matrix core must be linked to an argument position in the embedded logical structure: a. If the matrix predicate is a control verb, this follows (7.29); otherwise, b. If the matrix predicate is not a control verb, then link the unlinked syntactic argument in the matrix core to the logical structure argument position of the privileged syntactic argument of the linked core.

An example of a matrix-coding as non-privileged syntactic argument (‘raising to object’) construction from Icelandic (Thráinsson 1979) is given in (7.42). (7.42) a.

b.

Jón-Ø tel-ur ´ı barnaskap s´ınum John-MsgNOM believe-3sgPRES in foolishness his Harald-ur haf-i tek-ið bók-in-a. að CLM Harold-MsgNOM have.SBJ-3sgPRES take-PSTP book-DEF-FsgACC ‘John believes in his foolishness that Harold has taken the book.’ Jón-Ø tel-ur Harald-Ø ´ı barnaskap s´ınum John-MsgNOM believe-3sgPRES Harold-MsgACC in foolishness his bók-in-a. haf-a tek-ið have-INF take-PSTP book-DEF-FsgACC ‘John believes in his foolishness Harold to have taken the book.’

The (a) sentence involves a finite complement and no matrix coding, whereas the (b) form shows the construction in question. In the Icelandic examples the adverbial phrase ´ı barnaskap s´ınum ‘in his foolishness’ modifies Jón telur ‘John believes’ and hence is a constituent of the matrix core. The fact that the accusative NP Harald ‘Harold’ occurs between it and telur ‘believes’ shows that Harald is in


fact a constituent of the matrix core in (b). It is clear, then, that in the (b) example the accusative NP in the matrix core is a semantic argument of the verb in the linked core. As in the construction discussed in the previous section, there must be a core argument position in the matrix core which cannot be filled by a semantic argument from the logical structure of the matrix predicate. Since these verbs are obviously not (M)-atransitive like seem, the explanation for this open syntactic slot must lie elsewhere. In Foley and Van Valin (1984) it was noted that there is a systematic relationship between the S-transitivity of a verb when it takes NP or clausal syntactic arguments and that when it functions as a complement-taking predicate in a core juncture; namely, its s-transitivity is reduced by one in core junctures. This is illustrated in (7.43). (7.43) a.

b.

Three core arguments → two Phil told Dana a story [3] → Phil told Dana to . . . [2] Kim promised Sandy a picture of Chris [3] → Kim promised Sandy to . . . [2] Two core arguments → one Eileen remembered her purse [2] → Eileen remembered to . . . [1] Beckie wants a new Porsche [2] → Beckie wants to . . . [1]

It appears, then, that not only is the S-transitivity of the linked core reduced by 1 but that of the matrix core is as well. This is the case only when the linked core is a semantic argument of the matrix verb in logical structure. In purpose constructions, e.g. Bill brought the book for Sam to read, the purpose clause [core] is not a semantic argument of the matrix verb, and consequently there is no reduction of the S-transitivity of the matrix core. It is necessary, therefore, to amend (7.33b) to reflect this. The revised template selection principles are given in (7.44). (7.44) a.

b.

c.

Syntactic template selection principle (revised formulation): The number of syntactic slots for arguments and argument-adjuncts within the core is equal to the number of distinct specified argument positions in the semantic representation of the core. Universal qualification of the principle in (a): The number of slots in a core is reduced by 1 if: 1. it is the matrix core in a non-subordinate core juncture in which the linked core is a semantic argument of the matrix verb, and 2. it is the linked core in a non-subordinate core juncture. Language-specific qualifications of the principle in (a): 1. All cores in the language have a minimum syntactic valence of 1. 2. Argument-modulation voice constructions reduce the number of core slots by 1. 3. The occurrence of a syntactic argument in the pre/postcore slot reduces the number of core slots by 1 (may override c1).

What happens with believe? If it followed the pattern in (7.43) and (7.44b1), then we should have Juan believed the story [2] → *Juan believed to . . . [1] as the only two patterns with this verb. But this is not the case. Rather, with believe

255

256


we have Juan believed the story [2] → Juan believed Carlos to . . . [2]. This is also true for the other verbs which license the construction in (7.37b), e.g. expect, consider and find. Hence the crucial property of the verbs in this construction is that they are exceptions to the general pattern in (7.43) and therefore to (7.44b1) and have one more syntactic argument position in their core than they should. The actual linking in this construction is the same as that in the other matrixcoding construction, and it is illustrated for the Icelandic example in (7.42b) in Figure 7.12; the adverbial phrase ´ı barnaskap s´ınum ‘in his foolishness’ is omitted. If the linking in the second core had been passive, then the undergoer bók- ‘book’ would have appeared in the open matrix core slot, yielding the Icelandic equivalent of ‘John believes the book to have been taken by Harold’, Jón telur bókina hafa verið tekið af Haraldi. SYNTACTIC INVENTORY


NP

NUC

CORE NP

PRED

NP

PRED

V Jón

NUC

V

telur Harald hafa

tekið bókina.

5

1 LEXICON

ACTIVE: 3sg

PSA:NOM

Actor

ACC

2

Actor

ACTIVE: INF

3

ACC

Undergoer

believe¢ (Jón-, [[do¢ (Harald-, ø)] CAUSE [BECOME have¢ (Harald-, bók-)]]) Figure 7.12 Linking from semantics to syntax in the Icelandic construction in (7.42b)

The syntax to semantics linking algorithm in (7.41) applies to matrix-coding constructions with believe/telja, expect, etc., in exactly the same way as to the ones with seem/virðast discussed in the previous section. This should come as no surprise, since the two constructions are basically the same in terms of linking. The linking from syntax to semantics for Chris was believed by Pat to have eaten the bagel is given in Figure 7.13 on the next page. As before, the numbers refer to the steps in (7.41). Each core contains an M-transitive verb, and therefore step 1 applies to both. Because the verb in the first core is passive, and Chris must be a macrorole argument and is not the actor of believe, the actor of believe, Pat, is in the peripheryCORE marked by the preposition by. In the second core, the verb is active, and therefore its privileged syntactic argument is an actor and the bagel is an undergoer. Step 2 is straightforward, and in step 3 it is possible to link Pat with the actor of believe and the bagel with the undergoer of eat.


SENTENCE

PARSER

CLAUSE CORE NP AUX

LEXICON

PP

CORE NP

PRED 1c1

V

Chris was believed by Pat

Actor 2

PRED V

Non-actor MR 4

CLM

NUC

NUC

1c1 Voice?--Passive ∴ PSA = Non-actor

PERIPH

Voice?--Active to have eaten the bagel. ∴ PSA = Actor

Actor

Undergoer

3 Actor

3 Undergoer

believe¢ (x, [do¢ (y, [eat¢ (y, z)]) & INGR eaten¢ (z)]) Figure 7.13 Linking from syntax to semantics in English matrix-coding construction

There was no argument position available for Chris in the logical structure of believe, and therefore step 4 comes into play. The actor of eat is the only unlinked argument position in the logical structure, and it is the argument that would be the privileged syntactic argument of eat, were it to occur as the main verb in a simple clause. By step 4 Chris may be linked to the actor of eat, which yields the correct interpretation of the sentence and satisfies the completeness constraint. Note that the label ‘non-actor MR’ under Chris in Figure 7.13 means that Chris is not the actor of believe; it does not mean that the NP cannot be interpreted as the actor of a different verb. The constructional schemas for the English matrix-coding constructions are given in Table 7.6 on the next page. There are two schemas here, as the crucial feature of the second construction is the violation of (7.44b1); the first construction follows it. As in control constructions, the syntactic pivot is in core 2, not core 1, and it is, following the hierarchy for English, the highest ranking core macrorole. The linking is specified as ‘default’, since it follows (7.41) without modification. These schemas would also work for the Icelandic constructions, with the exception that there is no clause linkage marker in the Icelandic matrix-coding constructions.

7.3.3

Core junctures and the domain of case assignment

The two types of non-subordinate core junctures we have discussed, control and matrix-coding constructions, interact with the case-marking rules proposed in chapter 4 in different ways.

257

258


Table 7.6 Constructional schemas for English matrix-coding constructions Construction : English matrix-coding as PSA construction Syntax: Juncture: core Nexus: coordination Construction type: serial verb [CL [CORE NP [NUC . . .] (PP)] CLM [CORE [NUC . . .] . . .] . . .] Unit template(s): Core 1: (7.44) Core 2: (7.44) PSA: syntactic pivot of Core 2 – variable syntactic pivot (4.15a,c2) Linking: default Morphology: CLM to Semantics : propositional attitude, perception, evidential Prag m at i c s: Illocutionary force: unspecified Focus structure: unspecified Construction : English matrix-coding as non-PSA construction Syntax: Juncture: core Nexus: coordination Construction type: serial verb [CL [CORE NP [NUC . . .] NP] CLM [CORE [NUC . . .] . . .] . . .] Unit template(s): Core 1: violates (7.44b1) Core 2: (7.44) PSA: syntactic pivot of Core 2 – variable syntactic pivot (4.15a,c2) Linking: default Morphology: CLM to Semantics : propositional attitude, perception, cognition Prag m at i c s: Illocutionary force: unspecified Focus structure: unspecified

There is no real problem with case marking in the matrix core of control constructions. The basic rules for case assignment in accusative constructions from chapter 4 are repeated in (7.45). (7.45) a. b.

Case assignment rules for accusative constructions: assign nominative case to the highest ranking macrorole argument. assign accusative case to the other macrorole argument.

In a sentence like They persuaded us to go to the party, the third-plural actor is the highest ranking macrorole and therefore nominative, and the first-plural undergoer is the other macrorole and therefore accusative. The interesting examples are those


259

involving matrix coding, such as the Icelandic example in (7.42b) and its English equivalent, John believes Harold to have taken the book. In Icelandic, Jón ‘John’ is nominative and Harald ‘Harold’ is accusative, and the same pattern holds in English if we replace the proper nouns with pronouns, i.e. he believes him to have taken the book. Jón and he are the actor arguments of telja and believe, but Harald and him are the actor arguments of taka and take. Thus the matrix core contains two actor arguments; how is their case to be decided? The logical structure of these sentences is given in (7.46). (7.46)

believe (x, [[do (y, Ø)] CAUSE [BECOME have (y, z)]])

Jón or he is the x argument, and Harald or him is the y argument. There is a very simple solution to the problem at hand: only Jón and he are arguments of telja or believe, while Harald and him are not, and therefore Jón and he are the highest ranking arguments of telja and believe and receive nominative case, following (7.45a). Harald and him are the other macrorole arguments and therefore receive accusative case. Note that there would have been a serious problem if (7.45b) had been ‘assign accusative case to the undergoer’, since the accusative NPs in these examples are not undergoers. In sentences like (7.37a) and (7.39a), the only macrorole in the matrix core is a semantic argument of the predicate in the linked core; since it is the only macrorole in the core, it counts as the highest ranking and gets nominative case. Thus, the case assignment rules proposed in chapter 4 for simple sentences can account for case marking in the matrix cores in these examples. In section 4.4, it was stated that the case-marking rules in (4.25) apply to direct syntactic arguments within the core or in the pre/postcore slot, and since only simple sentences were discussed there, this was the only possible domain they could apply in. However, in core junctures there is more than one core in a clause, and so the question arises, do the case-marking rules apply to each individual core separately, or do they apply to all of the cores jointly within the clause? It turns out that languages vary with respect to the domain of case assignment: in some it is the clause, while in others it is the core. Icelandic presents the clearest example of a language in which the domain of case marking is the core. The case-assignment rules for Icelandic are given in (7.45). That the core is the domain of case assignment in Icelandic can be seen most readily in sentences like (7.47). (7.47)

Jón-Ø tel-ur mér (´ı barnaskap s´ınum) haf-a alltaf John-MsgNOM believe-3sgPRES 1sgDAT (in foolishness his) have-INF always ´ þo´ tt Olaf-ur leið inleg-ur. think.PSTP Olaf-MsgNOM boring-MsgNOM ‘John believes me (in his foolishness) to have always considered Olaf boring.’

What is crucial about this example is the occurrence of two nominative NPs, one in each core. If the core is the domain for the application of the case-marking

260


rules in (7.45), then the case pattern in (7.47) is accounted for, because Jón is the ´ highest ranking macrorole in the matrix core and Olafur is the highest ranking macrorole in the second, linked core. If the clause were the domain, then only one nominative NP would be possible, namely the highest ranking macrorole in the matrix core, Jón; all other macrorole arguments would be assigned accusative case, which they are not. In English, on the other hand, because the clause is the domain of case assignment only the highest ranking macrorole in the matrix core can be nominative; all other macroroles are accusative. This is illustrated in (7.48). (7.48) a. b.

Pat believed her to have told him to ask us to help them. For her to hire them would shock us.

The (b) example is particularly interesting, because it contains no nominative NP at all. This is because the highest ranking macrorole argument, the actor, is realized by an infinitival core, for her to hire them, and cores functioning as arguments do not carry case in English and many other languages. In Huallaga Quechua (Weber 1989), on the other hand, embedded cores and clauses do carry accusative case. Thus, languages vary with respect to the domain of case assignment: in Icelandic the rules apply independently in each core, while in English they apply to all of the cores in a clause jointly.

7.4

Linking in complex noun phrases

The primary issue regarding linking in complex NPs concerns relative clauses, in particular, the linking of the head noun to both the matrix clause and to the relative clause, since it functions in both. There are two main types of relative clauses, head-external and head-internal, and each presents a different linking problem: with head-external relatives, the problem is determining the function of the head inside the relative clause, whereas with head-internal relatives, the problem is determining which argument or adjunct in the relative clause also functions in the matrix clause. Within the class of head-external relatives, the two main types are those which have a relative pronoun, such as in English relative clauses with who or which, and those which have no relative pronoun and a gap in the relative clause, as exemplified by English sentences like The man (that) I saw is a spy. Linking in internally headed relative clauses will not be explicated here; see Van Valin and LaPolla (1997: section 9.3). The discussion will begin with the most common type cross-linguistically, externally headed relative clauses with no relative pronoun. Both English and Malagasy have this type of relative clause, but Malagasy has an extra twist: there is a restricted neutralization with respect to the head noun of the relative clause (Keenan 1976), while English lacks such a restriction.

7.4 Linking in complex noun phrases

(7.49) a.

Na-hita ny vehivavy (izay) nan-asa ny zaza Rakoto. PRFV.ATV-see DET woman CLM PRFV.ATV-wash DET child Rakoto ‘Rakoto saw the woman that washed the child.’ *‘Rakoto saw the woman that the child washed.’ ny zaza (izay) nan-asa ny vehivavy Rakoto. a . Na-hita PRFV.ATV-see DET child CLM PRFV.ATV-wash DET woman Rakoto ‘Rakoto saw the child that washed the woman.’ *‘Rakoto saw the child that the woman washed.’ ny zaza (izay) sas-an’ny vehivavy Rakoto. a . Na-hita PRFV.ATV-see DET child (CLM) wash-PASS-DET woman Rakoto ‘Rakoto saw the child that was washed by the woman.’ b. Trevor talked to the woman (that) Colin introduced him to.

b .

Trevor talked to the woman *(that) introduced Colin to him.

In the first two of the Malagasy examples, the head noun, which precedes the relative clause, can only be interpreted as the privileged syntactic argument of the relative clause; since Malagasy is an accusative language and the voice of the verb is active, it is always interpreted as the actor. In the third example it is interpreted as the undergoer, because the voice of the verb in the relative clause is passive. The only restriction that English has on this construction is that if the head noun is the privileged syntactic argument of the relative clause, as in (b ), then the clause-linkage marker that is obligatory; otherwise it is optional. The logical structure for (7.49a) is given in (7.50a), while the one for (7.49a ) is given in (7.50b). (7.50) a. b.

vehivavyi , [do (xi , [wash (xi , zaza)])])]) see (Rakoto, [be (---------- see (Rakoto, [be (-----zazai , [do (vehivavy, [wash (vehivavy, yi )])])])

Following the convention introduced in section 2.3, the head noun in the complex nominal logical structure is indicated by the thick dashed underlining. The underlining indicates that vehivavy ‘woman’ will be interpreted as the argument of -hita ‘see’, not the entire logical structure, in (a); the same holds for zaza ‘child’ in (b). Restrictive relative clauses are attributive modifiers, and accordingly they will be represented in an attributive logical structure, be (x, [pred ]), with the logical structure of the relative clause filling the ‘pred ’ slot in the attributive logical structure. While this is not a control construction, the same mechanism may be used for representing the function of the head noun within the logical structure of the relative clause; the head noun is coindexed with a lexically unfilled variable in the logical structure of the verb in the relative clause. In the linking from semantics to syntax, the head of the relative clause must be the privileged syntactic argument of the relative clause in Malagasy, and therefore if the head noun had been coindexed with a variable that would function as a non-actor, then passive or one of the other Malagasy voices would be necessary, as in (7.49a ). The linking from semantics to syntax in (7.49a) is illustrated in Figure 7.14.

261

262


SENTENCE

4

SYNTACTIC INVENTORY

CLAUSE CORE

NUC

NP

NP

COREN NUC N CLM

PERIPHERY N CLAUSE

PRED

CORE NUC

V N

NP

PRED V

Nahita ny vehivavy (izay)

nanasa

ny zaza Rakoto.

5 ACTIVE PSA:NOM

1 LEXICON

Actor

ACC

Undergoer

3

ACTIVE

2

ACC

Undergoer

see¢ (Rakoto, [be¢ (vehivavyi , [do¢ (xi , [wash¢ (xi, zaza)])])]) Figure 7.14 Linking from semantics to syntax in Malagasy relative clause in (7.49a)

The core template in the relative clause is missing a core argument position, one corresponding to the head noun, and so (7.44) must be revised to handle this. It already refers to syntactic arguments occurring in the precore slot, and this would account for relative clauses with relative pronouns. For other types of externally headed relative clause, however, there is no element in the precore slot, and therefore technically (7.44c3) does not apply to them. Since it is a universal feature of externally headed relative clauses that the core is missing an argument position when the head noun is a semantic argument of the verb or predicate in the relative clause, the relative clause provision to (b) should be added, rather than (c) in (7.44). Nothing needs to be said regarding cases in which the head noun is not an argument of the verb in the relative clause, because peripheral constituents are always optional in the syntactic templates. The revised version is given in (7.51). (7.51) a.

b.

Syntactic template selection principle (final formulation): The number of syntactic slots for arguments and argument-adjuncts within the core is equal to the number of distinct specified argument positions in the semantic representation of the core. Universal qualification of the principle in (a): The number of slots in a core is reduced by 1 if:


c.

1. it is the matrix core in a non-subordinate core juncture in which the linked core is a semantic argument of the matrix verb, and 2. it is the linked core in a non-subordinate core juncture; 3. it occurs in an externally headed relative clause construction in which the head noun is a semantic argument of the predicate. Language-specific qualifications of the principle in (a): 1. All cores in the language have a minimum syntactic valence of 1. 2. Argument-modulation voice constructions reduce the number of core slots by 1. 3. The occurrence of a syntactic argument in the pre/postcore slot reduces the number of core slots by 1 (may override c1).

The main complication which relative clauses introduce to the linking from syntax to semantics is that when the relative clause is recognized, an attributive logical structure must be introduced into the argument position occupied by the head noun, with the head noun functioning as the first argument of it, and the logical structure of the verb in the relative clause filling the ‘pred ’ slot in it. Since there will be an unlinked argument position in the semantics after all of the NPs in the clause are linked, the head must be linked to this position, in order to satisfy the completeness constraint. Hence it will be necessary to add a construction-specific condition to the linking specification in the constructional schema for relative clauses to deal with these additional complexities; this would be a general condition which all constructional schemas for externally-headed relative clauses would have. It can be formulated as in (7.52). (7.52) a. b. c.

d.

Conditions governing linking from syntax to semantics in externally headed relative clauses Retrieve from the lexicon an attributive logical structure and substitute the logical structure of the verb in the relative clause for the second argument. Link the head noun to the first argument in the attributive logical structure. Coindex the head noun in the attributive logical structure with either the unlinked argument position in the relative clause logical structure, or, if there is a relative pronoun, to the argument position linked to the relative pronoun. Insert the attributive logical structure into the argument position in the matrix logical structure occupied by the head noun, underlining the head noun.

Provision for externally headed relative clauses with relative pronouns (see (7.53) below) is made in (7.52c). The steps in (7.52) would come into play as soon as a relative clause is encountered; the steps in (7.41) apply within the relative clause just as within the matrix clause. The linking from syntax to semantics in (7.49a ) is given in Figure 7.15. Since Malagasy has a rich voice system, the first step is to determine the voice of the verb in the main clause; since it is active, we may conclude that the privileged

263

264

linking s yntax and semantics: complex sentences SENTENCE

PARSER

CLAUSE CORE NUC

NP COREN

NP PERIPHERYN

NUC N CLM

CLAUSE CORE

PRED

NUC V

N

NP

PRED

1c1

1c1

V

Main verb: (izay) Voice?−Active Nahita ny zaza ∴ PSA = Actor Undergoer (7.52b) 3

sasan'ny vehivavy Rakoto. Actor Actor

Actor Undergoer

Relative clause verb: Voice?−Passive ∴ PSA = Undergoer

2

[be¢ (x, [do¢ (y, [wash¢ (y, z)])])]

(7.52a) LEXICON

Actor Undergoer LEXICON

see¢ (v, w)

2

Coindexing (7.52c)

RC LS substitution (7.52d) see¢ (Rakoto, [be¢ (zaza i , [do¢ (vehivavy, [wash¢ (vehivavy, z i )])])])

Figure 7.15 Linking from syntax to semantics in Malagasy relative clause in (7.49a )

syntactic argument is the actor and that the NP immediately following the nucleus is the undergoer. With respect to the relative clause, the voice is passive, which means that the privileged syntactic argument is the undergoer and that the NP immediately following the nucleus is the actor. The head noun, ny zaza ‘the child’, is linked both to the second argument position in the logical structure of the matrix verb –hita ‘see’ and to the first argument position in the attributive logical structure containing the logical structure of the relative clause. Following (7.52c), the head noun in the first argument position in the attributive logical structure is coindexed with the unlinked argument in the relative clause logical structure. The logical structures of the main and relative clause verbs are related only indirectly through the common linking to the head noun. In the final step the w variable in the matrix logical structure is replaced with the attributive logical structure after the coindexing mandated by (7.52c), yielding the logical structure in (7.50b). The constructional schema for Malagasy relative clauses is given in Table 7.7.


Table 7.7 Constructional schema for Malagasy relative clause constructions Construction : Malagasy relative clause construction Syntax: Juncture: nuclearN Nexus: subordination (peripheral) Construction type: clausal modifier Unit template(s): main clause: (7.51) relative clause template: external head, [–PrCS], (7.51b2) PSA: head noun = variable syntactic pivot of relative clause (4.15a,c1) Linking: semantics → syntax – if actor = PSA, then marked voice syntax → semantics – (7.52) Morphology: CLM izay (optional) Semantics : restrictive modifier; be (xi , [pred ( . . . yi . . .)]), where y is lexically unfilled Pragmati cs: Illocutionary force: none (outside of potential focus domain) Focus structure: all elements are non-focal

The logical structure of relative clauses with relative pronouns differs from those discussed above only in that instead of the logical structure of the verb in the relative clause containing a lexically unfilled variable, the WH-word fills that position and is coindexed with the head. An example of this type of relative clause from English and its logical structure are given in (7.53). (7.53) a. b.

Pat liked the puppies which Kim saw. puppiesi , [see (Kim, whichi )])]) like (Pat, [be (---------

The linking from semantics to syntax follows the linking algorithm in (7.1), and the relative pronoun is linked to the precore slot, just as a WH-word is in a WH-question. The linking for (7.53a) is given in Figure 7.16 on the next page. The linking from syntax to semantics parallels that for the Malagasy example, with the exception that instead of linking the head noun with an unfilled argument position in the logical structure of the relative clause, it is coindexed with the WH-word in it. This is given in Figure 7.17 also on the next page. The relationship between the two main logical structures is expressed by the coindexing required by (7.52c). In order to integrate the two logical structures, the first argument in the attributive logical structure must be coindexed with the argument in the embedded logical structure containing the relative pronoun, following (7.52c). The whole attributive logical structure then fills the matrix verb logical structure argument variable which was coindexed with the relative pronoun by (7.52c). The result of the linking, when all of the logical structures are integrated, is (7.53b). Several crucial steps in the linking are supplied by the construction-specific linking requirements in (7.52), which are stated in the constructional schema for English relative clauses given in Table 7.8 on p. 267.

265

266


SENTENCE

4

SYNTACTIC INVENTORY

CLAUSE CORE NP

NP

NUC

COREN

PRED

PERIPHERYN

NUC N

V

CLAUSE CORE

PrCS

N

NP

NP

NUC PRED

PROREL

V

liked the puppiesi whichi Kim

Pat

saw.

5 1 LEXICON

ACTIVE: PSA:NOM 3sg

Actor like¢

ACC ACTIVE: PSA:NOM ACC 3sg

Undergoer

3

Actor Undergoer

(Pat, [be¢ (puppiesi , [see¢

(Kim,

2

which i)])])

Figure 7.16 Linking from semantics to syntax in English relative clause in (7.53a) SENTENCE PARSER

CLAUSE CORE NP

NUC PRED

NP COREN PERIPHERYN CLAUSE

NUC N

V PrCS N 1c1

NP PROREL

CORE NP

NUC PRED

1c1

V Relative clause Main verb: verb: Voice?--Active Pat liked the puppies which Kim saw. Voice?--Active ∴ PSA = Actor ∴ PSA = Actor Non-actor Actor 3 Actor MR 3 6 (7.52b) 2 Actor Undergoer Actor Undergoer 2 like¢ (v, w) LEXICON be¢ (x, [see¢(y, z)]) LEXICON RC LS substitution (7.52d) (7.52a) Coindexing (7.52c) like¢ (Pat, [be¢ (puppiesi, [see¢ (Kim, whichi)])])

Figure 7.17 Linking from syntax to semantics in English relative clause in (7.53a)

7.5 Reflexivization in complex sentences

Table 7.8 Constructional schema for English relative clause constructions Construction : English relative clause construction Syntax: Juncture: nuclearN Nexus: subordination (peripheral) Construction type: clausal modifier Unit template(s): Main clause: (7.51) relative clause template: external head relative clause: [±PrCS], (7.51b2) PSA: none Linking: syntax → semantics – (7.52) Morphology: CLM that (required if there is no relative pronoun and if head noun = PSA of subordinate clause; otherwise optional) Semantics : restrictive modifier; be (xi , [pred ( . . . yi . . .)]), where y is either a relative pronoun or lexically unfilled Pragmati cs: Illocutionary force: none (outside of potential focus domain) Focus structure: all elements are non-focal

7.5

Reflexivization in complex sentences

In section 5.2 a set of principles was presented which govern reflexivization in simple sentences, i.e. constructions with a single nucleus in a single core in a single clause. They are repeated in (7.54). (7.54) a.

Role hierarchy condition on reflexivization: (=(5.20)) the reflexive pronoun must not be higher on (4.14) (as applied to selection of privileged syntactic arguments in the language in (4.15c,d)) than its antecedent.

b.

Logical structure superiority (ls-superiority) (=(5.27)) A constituent P in logical structure is ls-superior to a constituent Q iff there is a constituent R in logical structure such that i. Q is a constituent of R, and ii. P and R are primary arguments of the same logical structure.

c.

Superiority condition on reflexivization: a bound variable may not be ls-superior to its binder.

d.

Domain of obligatory reflexivization constraint: (=(5.29)) one of two coreferring semantic co-arguments within a simple clause must be realized as a reflexive, while one of two coreferring syntactic arguments (which are not semantic co-arguments) within a simple clause may be realized as a reflexive.

Complex sentences present challenges to the hierarchy and domain conditions. With respect to the first issue, nuclear junctures create composite logical

267

268


structures; do all of the arguments in the derived logical structures behave like those in the logical structures of simple verbs? See Van Valin and LaPolla (1997: section 9.4.1), for detailed discussion. Matrix-coding constructions treat semantic arguments from different verbs as syntactic co-arguments; is reflexivization obligatory or optional in this instance? With respect to the domain condition, it is well known that some languages allow what is known as ‘long-distance reflexivization’, as illustrated in the following Icelandic example (Thráinsson 1991). (7.55)

Jón-Øi sag-ð -i að e´ g hef-ð -i John-MsgNOM say-PAST-3sg CLM 1sgNOM have.SBJ-PAST-1sg sigi . svik-ið betray-PSTP SELF ‘Johni said that I had betrayed himselfi .’

The ungrammaticality of the English translation indicates that this construction is not found in English. Core junctures raise a number of important questions. English does not appear to allow reflexivization across a core boundary within a clause containing a core juncture, but other languages do, as the following example from Icelandic (Maling 1986) illustrates; note again the ungrammaticality of the English translation. (7.56)

Harald-uri skipa-ð -i mér að rak-a sigi . Harold-MsgNOM order-PAST-3sg 1sgDAT CLM shave-INF SELF ‘Haroldi ordered me to shave himselfi .’

In (7.56) the controller and the reflexive are in different cores in the same clause. This seems to show that ‘within a simple clause’ in (7.54d) should be replaced by ‘within a core’. Control constructions of the type illustrated in (7.57) in English and other languages pose no problems for (7.54), however, since the controller and the reflexive are part of the same logical structure. (7.57) a. b.

Max persuaded himself to call Dana. Sally persuaded Tom to perjure himself.

The undergoer of persuade is a semantic argument of both verbs, and therefore it is a semantic co-argument of Max in (a) and of himself in (b). Hence both of these sentences meet the condition in (7.54d). A construction which does present a problem for (7.54d) is reflexivization in the matrix-coding construction, as illustrated in (7.58). (7.58) a. b.

Laurai believed herselfi /*heri to have been elected treasurer. Migueli believes himselfi /*himi to be the heir to the Spanish throne.

Even though Laura and herself are arguments of different logical structures, the role hierarchy condition as formulated in (7.54a) applies naturally here, since the actor of the matrix verb is the highest ranking argument in the logical structure.


Consequently, *Herself believed Laura to have been elected treasurer is ruled out; since herself in this sentence is the actor of the matrix logical structure and Laura is the undergoer in the embedded logical structure, the reflexive is higher on (4.14) than the controller, violating (7.54a). On the other hand, reflexivization is obligatory in this construction, and yet the antecedent and the reflexive are not semantic co-arguments; Laura and Miguel are semantic arguments of believe, and the reflexive is a semantic argument of the embedded logical structure. Accordingly the problem is with the revised version of (7.54d). Replacing ‘semantic co-arguments within a core’ with ‘syntactic co-arguments within a core’, for example, would work fine for the sentences in (7.58) and (7.57a), but it would not work for (7.57b), since Tom and himself are in different cores, or for the sentences in (7.59), in which the antecedent and the reflexive are also in different cores. (7.59) a. b.

Tanishai seems to have injured herselfi /*heri . Hamidi was believed to have recognized himselfi /*himi in the picture.

If the restriction were changed to ‘syntactic co-arguments within a clause’, which would cover all of these cases, then there would be no explanation for the ungrammaticality of the English equivalent to (7.56), *Harold ordered me to shave himself. The primary difference between this sentence and the ones in (7.57b) and (7.59) is that the antecedent and the reflexive are not semantic coarguments in *Harold ordered me to shave himself, while they are semantic coarguments in (7.57b) and (7.59). This strongly suggests that in English the domain restriction on the reflexivization of semantic co-arguments is different from the domain restriction on the reflexivization of co-referring syntactic arguments which are not semantic co-arguments. Thus, the domain restrictions seem to be as in (7.60). (7.60) a. b.

Domain restrictions on obligatory reflexivization in English: Co-referring semantic co-arguments: can be in different cores within a clause Co-referring syntactic co-arguments which are not semantic co-arguments: cannot be in different cores within a clause (one may be in PrCS with co-argument in the adjacent core)

This contrast falls out from the linking algorithms for complex sentences, in particular the syntax to semantics linking algorithm in (7.2). Semantic coarguments are by definition part of the same logical structure in the semantic representation of the sentence, and therefore it is possible to recover their semantic co-argumenthood across core boundaries in non-subordinate core junctures but not across clause boundaries, since clauses link independently of each other. In matrix-coding constructions like (7.58), the controller in the matrix core is linked to an argument position in the same logical structure as the reflexive; hence, even if they are in different cores, the controller and the reflexive will be linked to argument positions in the same logical structure. In control constructions like

269

270


(7.57b), in which the controller and the reflexive are in different cores, the theory of control in (7.29) links the controller in the matrix core to an argument position in the same logical structure as the reflexive. This is not the case in *Harold ordered me to shave himself, however; Harold is a semantic argument of order and himself of shave. They are not semantic co-arguments and are not in the same core, and therefore the sentence is ungrammatical. Because the obligatory sharing of a semantic argument in non-subordinate core junctures is the basis of the semantic co-argumenthood across a core boundary, this analysis predicts that reflexivization across core boundaries should not be possible in core subordination, due to the lack of any argument sharing, and this is correct, as (7.61) shows. (7.61) a. b.

*Danai regretted Bob’s kissing herselfi . *Debrai wanted very much for Sam to kiss herselfi .

It may be predicted, then, that the clause will universally be the syntactic domain for obligatory reflexivization of semantic co-arguments. On the other hand, there is nothing in the linking system that would constrain the interpretation of syntactic (co-)arguments which are not semantic co-arguments, and therefore it may be predicted that languages will vary quite substantially with respect to the treatment of syntactic arguments of this type. This is exactly what is seen in (7.56): in Icelandic coreferring syntactic co-arguments which are not semantic co-arguments can have a domain larger than a core, while in English they cannot. Indeed, Icelandic reflexivization is not even restricted by clause boundaries in some cases, as (7.55) shows. It is necessary, then, to reformulate the domain of obligatory reflexivization constraint for English in (7.54d) as in (7.62). (7.62) a.

b.

Obligatory reflexivization constraint in English: For semantic co-arguments, the domain of obligatory reflexivization is the clause: one of two coreferring core arguments which are semantic co-arguments must be realized as a reflexive. For coreferring syntactic co-arguments that are not semantic co-arguments, the domain of possible reflexivization is the core (and the precore slot): (1) if they are both direct arguments, then one of them must be realized as a reflexive; (2) if the lower ranking one in terms of (7.54a) is an argument-adjunct, then it may optionally be realized as a reflexive, subject to semantic conditions.

The condition in (a) accounts for the sentences in (7.57) and (7.59); the condition in (b1) accounts for the obligatory reflexives in (7.58), while the one in (b2) accounts for the pattern of reflexivization in (5.19b,c). The semantic condition of affectedness, proposed by Kuno and discussed in section 5.3, accounts for the impossibility of the reflexive in (5.19b) and its possibility in (5.19c). Thus, it appears that English does in fact have a type of long-distance reflexivization,


namely reflexivization involving semantic co-arguments across a core boundary. It does not, however, have long-distance reflexivization involving syntactic arguments which are not semantic co-arguments, unlike Icelandic and many other languages. There is a group of matrix-coding examples which raise a rather different problem for the principles in (7.54); they are given in (7.63) along with simplified semantic representations. John seems to himself to be sick. (7.63) a. a . TNS PRES [seem (himselfi , [sick (Johni )])] b. John is certain to appear to himself to be sick. b . TNS PRES [certain (appear (himselfi , [sick (Johni )]))] c. *It seems to himself that John is sick. c . TNS PRES [seem (himselfi , [it, TNS PRES [sick (Johni )]])]

The sentences in (7.63a,b) are interesting for two reasons. First, they have been cited as evidence against semantic accounts of reflexivization and in favor of syntactic accounts involving movement and traces, and second, native speakers have quite variable reactions to them, with some finding them fine and others finding them ungrammatical. Is it possible to account for both types of native speaker reactions within the framework of (7.54)? At first glance it might appear that (7.63a,b) violate the role hierarchy condition in (7.54a), since the reflexive seems to be higher on (4.14) than the controller. This, however, is an illusion. As argued in section 5.3, the hierarchy for English is actor > undergoer > other, and in these examples John is an undergoer and himself is a non-macrorole oblique core argument. Consequently, the binding is ‘undergoer binds other’, which follows (7.54a). The problem lies in (7.54c), the superiority condition: the reflexive appears to be l s-superior to its controller. This is not necessarily the case, however, and herein lies the crux of the different native speaker reactions. For those who find these sentences ungrammatical, the interpretation of (7.63a,b) with respect to (7.54b,c) is that because (1) himself is an argument of the matrix predicate seem and (2) John is the single argument of the embedded predicate sick , himself is ls -superior to John, and the binding violates the superiority condition. There is, however, another way to view the linking in these sentences. Because these are matrix-coding constructions, the embedded predicate sick and its argument John link independently and not as a unit. Hence for linking purposes John is not treated as part of the second argument of seem. This is illustrated in Figure 7.18 on the next page. This linking contrasts sharply with that in (7.63c), in which John and sick link as a unit, because the nexus is subordinate. This is illustrated in Figure 7.19.4 In this sentence John is sick links as a single unit, and therefore the superiority condition is unambiguously violated, rendering the sentence ungrammatical. Thus, speakers 4

See Van Valin and LaPolla (1997: section 9.1.1), for discussion of the semantic representation of sentences involving extraposition.

271

272



CLM PP

NUC

NP

NUC AUX PRED ADJ

PRED V John

CORE

seems to himself to be

sick. Undergoer

〈 TNSPRES [seem¢

(himselfi, [sick¢ (Johni )])]〉

Figure 7.18 Linking from semantics to syntax in (7.63a)


NUC

CORE

PP NP

PRED

NUC

AUXPRED ADJ

V It

CLAUSE

CLM

seems to himself

that John is

sick.

Undergoer Undergoer 〈

TNS

PRES [seem¢ (himselfi , [it, 〈

PRES [sick¢ (John )]〉])]〉

TNS

i

Figure 7.19 Linking from semantics to syntax in (7.63c)

who find (7.63a,b) grammatical are interpreting the linking in such a way that the superiority condition is not violated, while those who reject them are construing it in such a way that the condition is contravened.

7.6

Constraints on linking in WH-questions and related constructions

In the discussion of linking in WH-questions in chapter 5, it was noted that the WH-word can be mapped into the syntactic representation in one

7.6 Constraints: WH-questions

of four ways: into the precore slot, as in English; into the postcore slot, as in Dhivehi (Cain and Gair 2000); into the normal position for an NP argument or PP adjunct, as in Lakhota; or into the core-internal focus position, as in Turkish. There was no discussion of any restrictions on the linking from syntax to semantics, because there seem to be few if any of them in simple sentences. The picture is very different in complex sentences. Consider the following examples involving complex NPs. Mulder believes that Scully hid the files. (7.64) a. a . What does Mulder believe that Scully hid? b. Mulder believes the rumour that Scully hid the files. b . *What does Mulder believe the rumour that Scully hid? c. Scully interviewed the witness who saw the alien spacecraft. c . *What did Scully interview the witness who saw?

It is grammatical in English to form a WH-question with the question word functioning as the undergoer of the embedded clause, when the embedded clause is an object complement, as in (a ). When the embedded clause is a noun phrase complement, as in (b), or a relative clause, as in (c), the result is very different, as (b ) and (c ) clearly show. Ross (1967) argued that NP complements and relative clauses share a common structural feature, namely, the subordinate clause is embedded within a complex NP with a lexical head noun, and it is this property which blocks question formation. It also blocks the formation of related constructions, namely topicalization and relativization, as illustrated in (7.65). Those files Mulder believes Scully hid. (7.65) a. a . *Those files Mulder believes the rumour that Scully hid. a . *The alien spacecraft Scully interviewed the witness who saw. b. The files which Mulder believes that Scully hid were actually in the trunk of his car. b . *The files which Mulder believes the rumour that Scully hid were actually in the trunk of his car. b . *The alien spacecraft which Scully interviewed the witness who saw is stored in an abandoned missile silo in North Dakota.

The explanation proposed for these restrictions is the general principle of subjacency (Chomsky 1973). This principle has undergone a number of reformulations over the past three decades, but the basic idea is still that movement transformations cannot move an element across particular structural configurations. The idea that subjacency violations like those in (7.64b ,c ) and (7.65a ,a ,b ,b ) are caused by a syntactic rule moving an element across more than one bounding node runs into difficulties in languages like Lakhota in which questions words do not appear in the precore slot but rather occur in the normal core-internal position for a corresponding non-WH element, i.e. in situ. This might lead one to expect that there

273

274


would be no subjacency violations in such a language, but this is incorrect, as the following Lakhota examples show. (7.66) a.

b.

c.

Wiˇchásˇa ki [[ˇsu´¸ ka wa¸ igmú ki Ø-Ø-yaxtáke] ki le] wa¸ -Ø-Ø-yá¸ ke yelo. man the dog a cat the 3sgU-3sgA-bite the this 3sgU-3sgA-see DEC ‘The man saw the dog which bit the cat.’ Wiˇchásˇa ki [[ˇsu´¸ ka wa¸ táku Ø-Ø-yaxtáke] ki le] wa¸ -Ø-Ø-yá¸ ke yelo. man the dog a 3sgU-3sgA-bite the this 3sgU-3sgA-see DEC ‘The man saw the dog which bit something.’ Wiˇchásˇa ki [[ˇsu´¸ ka wa¸ táku Ø-Ø-yaxtáke] ki le] wa¸ -Ø-Ø-yá¸ ke he? man the dog a 3sgU-3sgA-bite the this 3sgU-3sgA-see Q ‘Did the man see the dog which bit something?’ *‘What did the man see the dog which bit?’

In (b) the undergoer of the relative clause has been replaced by táku ‘what, something (specific)’, and the result is a sentence with an indefinite-specific undergoer. In (c), the crucial example, the sentence has the question particle he and must be interpreted as a question. In section 5.4.3 it was mentioned that simple Lakhota sentences with táku and the question particle he are ambiguous between WHquestion and yes–no question interpretations, as (5.40d) showed; it is repeated below. (7.67)

ˇ u¸ ka ki táku Ø-Ø-yaxtáka S´ he? dog the 3sgU-3sgA-bite Q ‘What did the dog bite?’, or ‘Did the dog bite something?’

What is striking about (7.66c) is that it is unambiguous, even though it might be expected to be ambiguous just like (7.67). It cannot have the WH-question interpretation, unlike (7.67); it can only have the yes–no question reading. Put another way, táku cannot be interpreted as a question word in (7.66c); it can only be interpreted as an indefinite-specific pronoun. This is a subjacency effect, just as in (7.64c ) in English; in both examples, it is impossible to form a WH-question if the WH-word functions as (in these cases) a semantic argument in a definite restrictive relative clause. This is a very important fact. The account in terms of subjacency crucially refers to the movement of WH-words and other elements across certain phrasestructure configurations; the result of this movement is a long-distance dependency between the WH-word and a syntactic ‘gap’ in an embedded clause which spans these configurations. English, which has been analysed as having syntactic movement rules due to the displacement of the WH-word to the precore slot, shows subjacency effects, as demonstrated in (7.64) and (7.65). Lakhota, which presents no prima facie evidence for the existence of displacement in its grammar, also shows subjacency effects, as in (7.66c). Hence languages with ‘movement’ show subjacency effects, and languages without ‘movement’ also show subjacency


effects. It may, therefore, be concluded that ‘movement’ (displacement) is irrelevant to the explanation of these subjacency effects. It is necessary to look for some other feature common to the grammars of both types of languages for the explanation. An important clue to what this feature could be comes from the Lakhota sentence in (7.67). In section 5.4.3, it was argued that the two interpretations of this sentence result from different construals of what the focus of the question is. That is, if the focus of the question is on táku, then it is interpreted as a question word, yielding the meaning ‘What did the dog bite?’ If, on the other hand, the focus is on another constituent, then táku must be construed as an indefinite-specific pronoun, resulting in the reading ‘Did the dog bite something?’ If the question word being the focus of the question is the crucial precondition for the WH-question interpretation, then it follows that the reason (7.66c) is not ambiguous is that it is impossible to interpret táku as the focus of the question, leaving only the yes–no question reading possible. Why should it be impossible to interpret táku as the focus of the question? In section 3.3 the notion of the potential focus domain was introduced, the part of the sentence in which focal elements can occur; the actual focus domain, where focal elements actually occur in a particular utterance, must be within the potential focus domain. With respect to (7.66c), if táku cannot be the focus of the question, then there is no possible utterance in which it is in the actual focus domain, and from this it follows that it must be outside the potential focus domain of the sentence. The idea that some types of embedded clause are outside the potential focus domain is one encountered originally in section 6.7 in the discussion of focus structure in complex sentences. In that section a general principle governing the extent of the potential focus domain in complex sentences was introduced in (6.50), taken from Van Valin (1993b, 1995); it is repeated in (7.68). (7.68)

The potential focus domain in complex sentences: The potential focus domain extends into a subordinate clause if and only if the subordinate clause is a direct daughter of (a direct daughter of) the clause node which is modified by the illocutionary force operator.

As discussed in section 6.7, there is no limit in principle to the number of direct daughters involved, and accordingly the specification in parenthesis should be considered to be recursive. In terms of cross-linguistic variation, there appear to be only two possibilities: the potential focus domain is restricted to main clauses only, in which case (7.68) is irrelevant to the language, or the potential focus domain can extend into the deepest subordinate clause in any sentence, as long as the condition is not violated. The application of the principle in (7.68) to the Lakhota examples in (7.66) begins with an analysis of the structure of the sentences. The structure of Lakhota relative clauses was given in (B) in Figure 6.26. In addition, two more

275

276


SENTENCE

SENTENCE CLAUSE CLAUSE

CLAUSE NP

CORE

CORE

NP

CLAUSE Hokíla etá thaló ki manúpi iyúka he? Wiháa ki úka wa igmú etá wiháyaxtake ki le wayáka he? CLAUSE (B) (A) CLAUSE IF CLAUSE IF SENTENCE SENTENCE SENTENCE CLAUSE PERIPHERY

CORE NP

PP CLAUSE

NP

P

Wiháa etá wótapi ehúha, wiya ki mní iwíhakíiu he?

CLAUSE

CORE (C)

CLAUSE

IF

SENTENCE Figure 7.20 The structure of Lakhota complex sentences in (7.66c) and (7.69)

complex sentence types were given in Figure 6.26; these structures are repeated in (7.69). (7.69) a.

b.

[Hokˇs´ıla etá¸ thaló ki ma-Ø-Ø-nú-pi] i-Ø-Ø-yúkˇca¸ he? boy some[-SPEC] meat the INAN-3A-steal-pl INAN-3sgA-think Q ‘Does he think some boys stole the meat?’ he? Wiˇchásˇa etá¸ Ø-wóta-pi eˇchúha¸ , w´¸ı ya¸ ki mn´ı i-Ø-w´ıcˇ ha-Ø-kiˇciˇcu man some 3A-eat-pl while woman the water INAN-3plU-3sgA-get.for Q ‘While some men were eating, did the woman get them water?’

Simplified representations of each construction were given in Figure 6.26; they are repeated in Figure 7.20. In (A), the embedded clause is a direct daughter of the clause modified by the illocutionary force operator, and, consequently, the constituents of the embedded clause are within the potential focus domain. In (B) and (C), on the other hand, the embedded clause is not a direct daughter of the clause modified by the illocutionary force operator, and therefore the constituents of the embedded clauses are outside the potential focus domain. This conclusion is independently confirmed by the distribution of the indefinite-non-specific articles and by the possible felicitous responses to yes–no questions in these same complex sentences (cf. (6.53)–(6.57)). The fact that the relative clause in B is outside the potential focus domain predicts the impossibility of a WH-question reading in (7.66c). The results of the application of (7.68) to the structures in Figure 7.20 is given in (7.70).


(7.70) a.

Summary of potential scope of he: potential focus domain (in boldface) [Hokˇs´ıla etá¸ thaló ki manupi] ´ iyukˇ ´ ca¸ he? boys some meat the steal think Q ‘Does he think some boys stole the meat?’

b.

Wiˇchásˇa ki [[ˇsu´¸ ka wa¸ igmú etá¸ wiˇcháyaxtake] ki le] wa¸ yá¸ ka he? man the dog a cat some bite the this see Q ‘Did the man see the dog which bit some cats?’

c.

[Wiˇchásˇa ki wóte] eˇchuha ´ ¸ , tha-w´ıcˇ u ki mn´ı ik´ıcˇ iˇcu he? man the eat while his-wife the water get.for Q ‘While the man was eating, did his wife get him water?’

Only in (7.70a) do the individual constituents of the embedded clause function as independent information units; in (7.70b,c), on the other hand, the embedded clauses as a whole count as single information units. The internal constituents of the embedded clauses in these sentences are not included in the potential focus domain. The explanation for the impossibility of the WH-question interpretation of táku when it is in a definite restrictive relative clause makes a specific prediction about object complements and adverbial clauses: words like táku should be construable as question words in object complements but not in adverbial clauses. These predictions are correct, as (7.71) shows. (7.71) a.

b.

[Tuwá thaló ki manú] iyúkˇca¸ he? who meat the steal think Q ‘Who does he think stole the meat?’, or ‘Does he think someone stole the meat?’ [Wiˇchásˇa ki táku yúte] eˇchúha¸ , tha-w´ıcˇ u ki mn´ı ik´ıcˇ iˇcu he? man the eat while his-wife the water get.for Q ‘While the man was eating something, did his wife get him water?’ *‘What did his wife get him water, while the man was eating?’

It is clear, then, that in order for táku, tuwá or one of the other question words/ indefinite-specific pronouns in Lakhota to be interpreted as a question word, it must occur in the potential focus domain of the sentence. This may be expressed in a preliminary fashion as the constraint on question formation in (7.72). (7.72)

Constraint on question formation (preliminary formulation): The element questioned (the question word in a simple, direct WH-question5 or the focal NP in a simple, direct yes–no question) must function in a clause into which the potential focus domain of the sentence extends.

It was shown in section 6.7 that the possible interpretation of the focus in a yes– no question is affected by the constraint in (7.68), and therefore the restriction in (7.72) applies to both types of questions. Thus, the principles in (7.68) and (7.72) provide an explanation for the observed subjacency effects in Lakhota. 5

This constraint does not apply to echo questions, rhetorical questions or metalinguistic questions.

277

278


Can this analysis be applied to languages like English, in which questions words do not occur in situ but rather in the precore slot? In languages of this type, the position of the WH-word in the question is not relevant to explaining the subjacency effects, because in all questions of this kind the WH-word occurs in the precore slot, regardless of the grammaticality of the question. Rather, what is relevant is whether the potential focus domain extends into the clause in which the question word functions semantically. In applying the principle in (7.68) to English in section 6.7, it was found that the potential focus domain extends into object complements (see Figure 6.24) but not into ad-core subordinate clauses (see Figure 6.25). If the principle in (7.68) is applied to the English relative clause in Figure 7.16, it is readily apparent that the embedded clause is not a direct daughter of the clause modified by the illocutionary force operator, and therefore relative clauses in English, as in Lakhota, function as a single information unit.6 Finally, given the structure of the noun complement clause in Figure 6.27, the embedded clause is within a NP and therefore could not be a direct daughter of the clause modified by the illocutionary force operator; accordingly, it too counts as a single information unit. These results may be summarized in Table 7.9. With respect to the examples in (7.64), the results in Table 7.9 together with the principle in (7.72) account for the grammaticality or ungrammaticality of all of the sentences. Forming a question in which the WH-word functions in an object complement, as in (7.64a ) is grammatical, as predicted, whereas forming a question in which the WH-word functions in a relative clause or noun complement clause, as in (c ) and (b ), respectively, is ungrammatical, again as predicted. It also explains the ungrammaticality of the English translation of the Lakhota example in (7.71b), where the WH-word functions in an ad-core subordinate clause. Table 7.9 Potential focus domain in English complex sentence constructions Construction

Structure represented

Direct daughter?

Potential focus domain?

Object complement Ad-core clauses Relative clause (rest.) Noun complement

Figures 6.17, 6.24, Figure 6.25 Figures 6.28, 7.16, Figure 6.27

Yes No No No

Yes No No No

An explanation has been provided for the subjacency effects in both English and Lakhota which does not require the existence of a long-distance dependency between a WH-word in the precore slot and a syntactic ‘gap’ in an embedded clause. Rather, what is common to the two languages is the crucial role of the potential focus domain in constraining question formation. This is captured in the principles in (7.68) and (7.72), which apply equally to both languages, despite their 6

Non-restrictive relative clauses would also be outside the potential focus domain, since they do not meet the structural condition in (7.68), and in addition, because they are embedded sentences, they have their own illocutionary force operator, which is obligatorily declarative (see Figure 6.29); they are a distinct focus domain from that of the matrix clause. Hence extraction should be impossible out of them, which is the case.


manifest syntactic differences. The principle in (7.72) can be integrated naturally into the linking algorithms. In order to block the generation of sentences like (7.64b ), it is necessary to modify step 5b of the semantics to syntax algorithm as in (7.73); the modification is in boldface. (7.73)

Linking algorithm: semantics → syntax (Revised) 1. Construct the semantic representation of the sentence, based on the logical structure of the predicator. 2. Determine the actor and undergoer assignments, following the actor–undergoer hierarchy in Figure 4.4. 3. Determine the morphosyntactic coding of the arguments a. Select the privileged syntactic argument, based on the privileged syntactic argument selection hierarchy and principles in (4.14)–(4.15). b. Assign the argument(s) the appropriate case markers and/or adpositions. c. Assign the agreement marking to the main or auxiliary verb, as appropriate. 4. Select the syntactic template(s) for the sentence following the principles in (7.51). 5. Assign argument(s) to positions in the syntactic representation of the sentence. a. Assign the [−WH] argument(s) to the appropriate positions in the clause. b. If there is a [+WH] argument in the semantic representation of a clause into which the potential focus domain extends, 1. assign it to the normal position of a non-WH-argument with the same function, or 2. assign it to the precore or postcore slot, or 3. assign it to a position within the potential focus domain of the clause (default = the unmarked focus position). c. A non-WH argument may be assigned to the precore or postcore slot, subject to focus structure restrictions (optional). d. Assign the arguments(s) or logical structure in logical structure(s) other than that of the predicator in the nucleus to 1. a periphery (default) a. If the representation is pred (NP/LS, LSMAIN ), where pred is a prepositional predicate, then assign the P + NP/Core/ Clause to the peripheryCORE . b. If the representation is LSMAIN pred /connective NP/LS, then assign the P + NP or CLM + Clause to the peripheryCLAUSE . 2. the precore or postcore slot, or 3. the left- or right-detached position.

The addition of the requirement ‘in the semantic representation of a clause into which the potential focus domain extends’ makes it impossible to link a WH-word to any position in the syntactic representation if this condition is not met, resulting in a completeness constraint violation. This condition in step 5b

279

280


requires that focus structure information be available to the linking system, as in Figure 5.35. In order to constrain the linking from syntax to semantics, it is necessary to modify step 6 in (7.2) as follows. (7.74)

Linking algorithm: syntax → semantics (revised formulation) 1. Determine the macrorole(s) and other core argument(s) in the clause. a. If the verb is intransitive, then assign the privileged syntactic argument either macrorole or direct core argument status, depending upon the language (language-specific) b. If the verb is transitive and the language lacks voice oppositions, determine the macroroles from case marking and/or word order (language-specific). c. If the language has a voice opposition, determine the voice of a transitive verb (language-specific): 1. If the construction is syntactically accusative: a. If it is the unmarked voice, the privileged syntactic argument is actor. b. If it is passive, the privileged syntactic argument is not the actor of the predicate in the nucleus; 1. the actor may appear as a direct core argument (language-specific); or 2. the actor may appear in the peripherycore marked by an adposition or an oblique case (language-specific); or 3. if there is no actor in the core or the periphery, then replace the variable representing the highest ranking argument in the logical structure with ‘Ø’. 2. If the construction is syntactically ergative: a. If it is the unmarked voice, the privileged syntactic argument is undergoer. b. If it is antipassive, the privileged syntactic argument is actor; 1. the undergoer may appear as a direct core argument or as an oblique element (language-specific); 2. If there is no undergoer in the core or the peripherycore , then replace the variable representing the lowest ranking argument in the logical structure with ‘Ø’. 3. assign macrorole status to the other direct core argument, if it is not dative or in an oblique case (language-specific). d. If the language is head-marking and there are independent NPs in the clause, associate each NP with a bound argument marker (language-specific). 2. Retrieve from the lexicon the logical structure of the predicate in the nucleus of the clause and with respect to it execute step 2 from (7.73), subject to the following proviso: a. If the language allows variable undergoer selection and if there is more than one choice for undergoer, do not assign undergoer to an argument in the logical structure. b. Determine the linking of the non-macrorole core argument:


1. If there is a two-place state predicate in the logical structure and if the non-macrorole core argument is marked by a locative adposition or dative or a locative-type case, then link it with the first argument position in the state predicate in the logical structure and link the other non-actor core argument (if there is one) to the second argument position in the state predicate, or 2. If there is a two-place state predicate in the logical structure and if the non-macrorole core argument is not marked by a locative adposition or dative or a locative-type case, then link it with the second argument position in the state predicate and link the other non-actor core argument (if there is one) to the first argument position in the state predicate. 3. Otherwise, link the animate NP with the first argument position in the state predicate in the logical structure. 3. Link the arguments determined in step 1 with the arguments determined in step 2 until all core arguments are linked. 4. In non-subordinate core junctures, one of the arguments of the matrix core must be linked to an argument position in the embedded logical structure: a. If the matrix predicate is a control verb, this follows (7.29); otherwise, b. If the matrix predicate is not a control verb, then link the unlinked syntactic argument in the matrix core to the logical structure argument position of the pivot of the linked core. 5. If there is an adjunct phrase in a periphery, a. and if it is in the peripheryCORE , then retrieve the logical structure of the predicative adposition from the lexicon, insert the logical structure of the core as the second argument in the logical structure and the object of the adposition as the first argument; b. and if it is in the peripheryCLAUSE , then link the adjunct PP or clause logical structure to the matrix logical structure via the semantic representation of the adposition or clause-linkage marker. 6. If there is an element in the pre- or postcore slot, or a WH-word in situ (language-specific), a. assign it the remaining unlinked argument position in the semantic representation of the sentence, provided that the logical structure to which it is linked is for a clause into which the potential focus domain extends. b. if there are no unlinked argument positions in the sentence, then treat the WH-word like a predicative preposition and follow the procedure in step 5. 1. Treat the entire logical structure of the sentence as the linked part of the prepositional or clause-linkage marker logical structure (default); or 2. If the potential focus domain extends into the embedded clause, then treat only the logical structure of the embedded clause as the linked part of the prepositional or clause-linkage marker logical structure (optional).

281

282


The constraint in step 6a prevents a WH-word from being linked to the semantic representation unless the condition in (7.72) is met. In a language like English it would leave the WH-word stranded in the precore slot, resulting in a completeness constraint violation. In a language like Lakhota, on the other hand, it would block the linking if táku is construed as a question word, likewise resulting in a completeness constraint violation. But there is another option in Lakhota; táku can also be interpreted as an indefinite-specific pronoun, and on this interpretation the condition in (7.72) does not apply, thereby allowing the linking of táku to a variable in the logical structure, satisfying the completeness constraint, and yielding the yes–no question reading. The reformulation of step 6b makes it possible to capture well-known constraints on the interpretation of adjunct WH-questions, as exemplified in (7.75). (7.75) a. b.

When did Skinner say that Krycek would be at the missile silo? When did Scully interview the witness who saw the alien spacecraft in the silo?

The sentence in (a) is ambiguous; it can be a question about when something was said or about when someone would be somewhere. The sentence in (b), however, is not ambiguous; it can only be a question about when the interview took place, not about when someone saw something. The issue here is again what can be questioned: more specifically, what can be questioned with respect to the time of its occurrence. It would be expected, then, that the potential focus domain would play a key role in constraining the interpretation of the question, and this seems to be the case. Example (7.75a) involves a structure which is compatible with the potential focus domain extending into the embedded clause, and accordingly it is ambiguous with respect to the interpretation of the scope of when. The structure in (b), however, does not meet the condition in (7.68) (cf. Table 7.9), and therefore the potential focus domain does not extend into the embedded clause; hence it cannot be the focus of the question, and when must be interpreted as modifying the matrix clause. Ad-clausal subordinate clauses have an interesting property not shared with ad-core subordinate clauses. They are like ad-core clauses in that extraction out of them is impossible, since they do not satisfy the principle in (7.68); they are a ‘sister’ of the clause node modified by the illocutionary force operator, not a daughter (see Figure 6.7). The impossibility of extraction out of them is illustrated in (7.76). (7.76) a. b. c.

*Who was John upset, because Mary kissed —? *Who would John cry, if Mary had been seeing —? *Who was Mary happy, even though John had kissed —?

However, like object complements, which are daughters of the matrix clause node, and unlike ad-core subordinate clauses, ad-clausal subordinate clauses do allow inversion; the following examples are from Haumann (1997).


(7.77) a. b. c. d. e. f.

She said that [under no circumstances] would she marry him. Max thinks that [because Tom left early] he was angry at him. He left early because [never again] does he want to end up completely drunk. Bill would cry if [while in London] Mary had been seeing John. *I haven’t talked to him since [when I least expected it] he got divorced. *He got ill while [in Italy] he was on holiday.

It has been recognized since at least Hooper and Thompson (1973) that inversion is a property of asserted clauses, and for an embedded clause to be asserted, it must be within the potential focus domain. The principle in (7.68) correctly predicts the grammaticality of (7.77a,b) and the ungrammaticality of (7.77e,f), but it fails to predict the grammaticality of (7.77c,d). Moreover, there is something of a paradox here: extraction is impossible out of these clauses, which follows from them being outside of the potential focus domain, while inversion is possible in them, which follows from them being within the potential focus domain. How can this apparent paradox be resolved? Bickel (1993) suggests that the potential focus domain (and therefore the actual focus domain) can be in either the main clause or the ad-clausal subordinate clause but not both in this construction. He calls this a ‘Rubin effect’, which is the famous phenomenon from Gestalt psychology whereby an ambiguous figure, e.g. a duckrabbit, can be seen as only one form or the other at any given instant. In order to extract out of an ad-clausal subordinate clause, the potential focus domain has to encompass both clauses, which he argues is impossible. In a normal extraction out of an embedded clause, the potential focus domain includes the whole structure, with the actual focus domain being in the embedded clause. Why should there be such a ‘Rubin effect’ with this construction? Why should it be impossible for the potential focus domain to encompass both clauses? A clue can be found in the semantic representation of these constructions, which is given in (6.38). An examination of the representations in (6.38) reveals that in all of the complex sentence types in which the potential focus domain can extend through the whole sentence, there is an embedding relationship among the logical structures. That is, in (6.38d–l) the linked logical structure is embedded in the matrix logical structure. If one assumes, reasonably, that each logical structure supplies the propositional content for a potential assertion, then when the linked logical structure is embedded in a higher logical structure, it contributes its propositional content to the entire representation, which may be expressed as a single speech act.7 Hence the potential focus domain can extend into the entire sentence, matrix plus embedded clauses. The representations in (6.38n–s) are quite different, however: each logical structure is independent of the other, and they are linked by some kind of connective. Consequently, logical structures are not 7

Direct discourse constructions in (6.38m) are not included here, because they have the unique property of allowing an independent illocutionary force operator on the embedded logical structure; hence the embedded logical structure always underlies a distinct speech act from the matrix logical structure.

283

284


part of a single complex proposition which can be expressed as a single speech act; rather, because they are independent of each other, they are potentially independent utterances. With respect to the representations in (6.38r,s), this is indeed the case; they are realized as conjoined clauses, each with its own independent illocutionary force operator. Reason, conditional and concessive clauses, on the other hand, are not linked by coordinating conjunctions but rather by subordinating conjunctions, each with their own substantial semantic content to contribute to the sentence. Because they are not daughters of the sentence node, they cannot have their own illocutionary force operator. This means that the semantic representation for these three types of ad-clausal subordinate clauses would be as in (7.78): (7.78)

IF DEC [LS1 ] connective [LS2 ]

Because there are two independent logical structures and only one illocutionary force operator, only one of the logical structures can be construed as asserted at a time. This is the ‘Rubin effect’. Inversion is possible in the ad-clausal subordinate clause because it is possible for the potential focus domain to extend into the clause instantiating either LS1 or LS2 . If this is the case, then why is extraction impossible out of the clause instantiating LS2 ? As stated earlier, in order for there to be extraction out of an embedded clause, the potential focus domain must extend into the whole sentence, and that is not possible in this construction. In this regard, the sentences in (7.76) are similar to conjoined constructions with respect to extraction, as (7.79) illustrates. (7.79) a. b.

John cleaned the garage, and Mary washed the dishes. *What did John clean the garage, and Mary washed ?

The overt marking of interrogative illocutionary force is in the first clause, while the WH-word is in fact related to the logical structure of a clause with a distinct illocutionary force operator, a clause whose form indicates declarative illocutionary force; accordingly, it is impossible to link the WH-word in the first clause to an argument position in the logical structure of the second clause. Likewise in (7.76), the interrogative marking is in the matrix clause and hence applies to the matrix clause only; the ad-clausal subordinate clause cannot be construed as interrogative, and thus the same linking problem exists as in (7.79b). Note that it is possible to have a rhetorical WH-question inside a because-clause. (7.80)

It’s good that Chris took on solving the problem, because who else is crazy enough to try it?

This could only be possible if the scope of the illocutionary force operator in (7.78) were over the clause instantiating LS2 . The discussion thus far has concentrated on WH-question formation, but, as shown in (7.65), topicalization and restrictive relativization are also subject to the same constraints. In many languages these constructions are syntactically very similar, but functionally they differ in terms of the discourse function of the


Table 7.10 Discourse functions of ‘extracted’ element in ‘extraction’ constructions Construction

Function of ‘extracted’ element

WH-question Topicalization Relativization

Focus Focus or Topic Topic

WH-word in a question, the non-WH NP in the precore slot in a topicalization construction, and the head of a relative clause. This is summarized in Table 7.10. There is an obvious connection between WH-elements, which are always focal, and focal non-WH elements in the precore slot, on the one hand, and the potential focus domain, on the other, and this is the basis for the provisions in the linking algorithms introduced in (7.73) and (7.74). Why should topicalization and restrictive relativization also be constrained by the potential focus domain, given that the NP in the precore slot in a topicalization construction can be topical and that the head noun is always topical with respect to the relative clause? The essential feature that these two constructions share is that the clause in which the displaced NP functions is always about the referent of the NP (Kuno 1987). The central notion, then, is pragmatic aboutness; the restrictive relative clause must be interpretable as being about its head, and the sentence fragment following a topical element in the precore slot must likewise be interpretable as being about the precore slot element. This condition may be formulated as in (7.81); Kuno (1987) presents a similar constraint. (7.81)

Pragmatic-aboutness condition on topicalization and relativization: The sentence fragment following a topical element in the precore slot or a restrictive relative clause must be pragmatically interpretable as being about the precore slot element or the head noun.

Reinhart (1981) presents an analysis of pragmatic aboutness. She argues that in order for a sentence to be about the referent of an NP, it must be possible to form an alternative sentence (with the same essential structure) in such a way that the NP at issue functions as the focus of a possible assertion that the sentence can be used to make. To illustrate what she means here, look again at (7.65a), Those files Mulder believes Scully hid. Reinhart’s analysis requires that, in order for the sentence fragment Mulder believes Scully hid to be construable as pragmatically about the precore slot NP those files, there must be an alternative form of this sentence in which the NP those files serves as the focus of an assertion. The sentence in (7.64a), Mulder believes that Scully hid the files, is such a sentence; how can it be shown that the files can serve as the focus of (7.64a)? As discussed in section 6.7, only the asserted part of an utterance can be interpreted as being negated, and, accordingly, if the constituent can be negated or denied in a conversational exchange, then it is a possible focus. This is illustrated in (7.82).

285

286


(7.82)

Speaker 1: Mulder believes that Scully hid those files. Speaker 2: No, the keys to his car.

The fact that this is a felicitous exchange shows that the files can be negated and therefore is a possible focus in (7.64a). This meets Reinhart’s criterion for pragmatic aboutness, and consequently it may be concluded that the sentence fragment Mulder believes Scully hid is interpretable as being about the NP those files in (7.65a). Hence the construction in (7.65a) meets the condition in (7.81) and is predicted to be grammatical, which it is. Her criterion makes crucial reference to the constituent serving as the focus of a possible assertion that the sentence can be used to make, and here is where the potential focus domain comes in: in order for a constituent to be the focus of a possible assertion that a sentence can be used to make, it must be within potential focus domain. Now compare (7.65a) with (7.65a ), *The alien spacecraft Scully interviewed the witness who saw. In order to determine whether the sentence fragment Scully interviewed the witness who saw can be interpreted pragmatically as being about the NP the alien spacecraft, it is necessary to take an alternative form of the sentence and ascertain if the alien spacecraft is a possible focus in it. The alternative sentence is (7.64c), Scully interviewed the witness who saw the alien spacecraft; it is necessary to place it in the same type of context as in (7.82) to see if this NP can be negated. (7.83) a. b.

Speaker 1: Scully interviewed the witness who saw the alien spacecraft. Speaker 2: *No, some lights in the sky. Speaker 1: Scully interviewed the witness who saw the alien spacecraft. Speaker 2: No, Skinner.(= ‘Skinner interviewed the witness’, = ‘Scully interviewed Skinner’, but not = ‘the witness who saw Skinner’)

The results are very different from (7.82). In both interchanges it is impossible to deny the alien spacecraft, and this means that it cannot be interpreted as the focus of a possible assertion that (7.64c) can be used to make. If that is the case, then the sentence fragment Scully interviewed the witness who saw cannot be interpreted pragmatically as being about the alien spacecraft, and therefore the condition in (7.81) is not met, and the topicalization construction in (7.65a ) is predicted to be ungrammatical, which it is. In order for a constituent to be negatable (deniable) as in (7.82), i.e. without repeating the whole previous utterance, it must be interpretable as the focus of the assertion, and this requires it to be in the actual focus domain. This in turn requires that the potential focus domain extend into the clause containing the constituent. Restrictive relativization and the occurrence of topical elements in the precore slot are constrained by the condition in (7.81), which crucially builds on Reinhart’s analysis of pragmatic aboutness. In her analysis, in order for a sentence fragment or relative clause to be about the referent of an NP, it must be possible to form an alternative sentence (with the same essential structure) in such a way that the constituent at issue functions as the focus of a possible


assertion that the sentence can be used to make. This requires the constituent to be in the potential focus domain. Thus, even if the displaced NP is a topic, the interpretation of the construction is still constrained by the potential focus domain. If it is focal, then the interaction with the potential focus domain is the same as for WH-questions. Thus, all of the constructions in Table 7.10 are constrained by the potential focus domain, and therefore the provisions in the linking algorithms regarding the potential focus domain will work for them, too. There is still one puzzle remaining regarding relative clauses. While it makes sense to talk about the sentence asserting something about the topic NP in the precore slot in a construction like (7.65a), for example, it is not possible to say the same thing about (7.65b), in which the files is the head noun and which Mulder believes Scully hid is the relative clause. Restrictive relative clauses do not assert anything; as restrictive modifiers, they are presupposed. Why, then, should the potential focus domain restrict the interpretation of constructions which are by definition not asserted? A plausible answer lies in the connection between assertion and predication. Cattell (1984) and Kluender (1992) argue that the constraints being investigated are derived from restrictions on the formation of complex predication structures. This goes back to the point made earlier with respect to the logical structure of ad-clausal subordination and related peripheral subordination structures in contrast to daughter subordination. When logical structures are embedded in a higher logical structure, the resulting complex representation can be interpreted as a single complex predication underlying a single speech act. This is true of (6.38d–l), all of which roughly have the structure ‘pred (x, [. . . pred (y, [. . . pred (z, . . .’, in which each logical structure is an argument of the logical structure in which it is embedded. This is not the case in (6.38n–s), in which the logical structures do not have this nested structure and do not form a single complex predication. The impossibility of extraction out of ad-clausal subordinate clauses would follow from the constraints that Cattell and Kluender propose. It has long been recognized that the subject–predicate opposition is fundamentally one of topic and comment, with the predicate being an assertion about the topic (see the discussion of predicate focus in section 3.2, also Lambrecht (1994:232)). If this is the case, then limits on what can be construed as an assertion about a topic are also limits on what can be construed as a possible predication about a ‘subject’. Restrictive relative clauses are complex predications modifying an NP; note that the logical structure of a head noun + relative clause is the same as that of an attributive construction. Thus the predication relationship between The boy and is tall in The boy is tall is analogous to that between the files and which Mulder believes that Scully hid in (7.65b). It is uncontroversial that the predicate is tall asserts something about the boy in The boy is tall, and there it is clear that predication and ‘assertion about’ are fundamentally related notions. If this is the case, then the application of constraints on ‘assertion about’ to predication follows naturally. Hence, the head noun must be related to the relative clause in such a way that if the relative clause were an independent sentence and the

287

288


head noun a topic, the relative clause could be construed as an assertion about that topic. And the precondition for this is that the NP serving as the head noun must function within the relative clause in such a way that if the relative clause were an independent assertion containing the NP, the NP would be in the potential focus domain. Thus, despite the different functions of the displaced elements in these three constructions, they are all ultimately constrained by whether the potential focus domain extends into the embedded clause in question, and this follows from (7.68). The constraint in (7.68) represents the default distribution of the potential focus domain in complex sentences. There are a number of other factors which interact with it to reduce or extend the potential focus domain. In particular, lexical semantic factors may also influence the potential focus domain, both in terms of preventing a position in the potential focus domain from being the actual focus domain and of overriding the principle in (7.68) and permitting the actual focus domain to be in structural configurations where it would otherwise be impossible. These two possibilities can be illustrated in English. The principle in (7.68) predicts that WH-question formation, etc., should be possible out of object complements, but this is not always true. It has long been noted that while it is very easy to form a question out of the complement of say, it is highly odd to do this out of the complements of verbs of manner of speaking, e.g. ?? What did Fred murmur/chortle/lisp that Mary had bought? There is a straightforward Gricean explanation for this (Grice 1975). The focus of an utterance is the most informationally rich part, and the selection of say, the most semantically neutral verb of saying, together with an unmarked intonation pattern, indicates that the primary information content of the utterance is the substance of the communication, which is syntactically expressed in the complement clause. Hence the focus can fall in the that-clause, making question formation, etc., possible. The choice of a verb which highlights the way in which something is said rather than what is said, such as murmur, chortle and lisp, causes the focus to shift to the verb in the main clause, because of the maxim of relevance: the speaker’s choice of an informationally richer expression (murmur) over another, more neutral possibility (say) only makes sense in terms of the cooperative principle if the manner of expression is in fact highly relevant to the main point of the utterance. Hence the focus must fall on the matrix verb, keeping the complement from being the actual focus domain despite the fact that the structure as a whole meets the condition in (7.68). The same thing occurs in ‘extraction’ out of NPs: when the main verb is not informationally distinctive, as in Who did you read a book about —?, forming a question out of the PP in the undergoer NP is fine; however, when the verb is informationally rich, it naturally draws the focus for the same Gricean reason as above, precluding the possibility of the object NP being the actual focus domain, e.g. *Who did you deface/lose/destroy a book about —? The second type of lexical semantic effect is exemplified with complex NPs like make the claim or hold the belief. The structure of a sentence like Fred made the claim that Mary stole the money does not meet the condition in (7.68), because


the subordinate clause is part of an NP and is not a direct daughter of the matrix clause node; hence the potential focus domain should not extend into it, and question formation out of it should be impossible. However, it has long been known that question formation is in fact possible for at least some speakers, e.g. What did Fred make the claim that Mary stole?, and it has usually been argued that this question is acceptable because make the claim that X is virtually synonymous with claim that X, an expression whose structure meets the principle in (7.68). When there is no simple object complement paraphrase, as in e.g. *What did Fred investigate the claim that Mary stole?, then the question is ungrammatical as predicted. In the make the claim examples, lexical semantic factors have overridden the principle in (7.68) to permit an otherwise excluded structure to fall within the actual focus domain. Finally, discourse considerations may also affect the interpretation of the focus domains, as Kuno (1987) has argued. For example, the odd extraction-from-NP question above can be made rather more acceptable if it is part of an exchange like the one in (7.84), from Kuno (1987). (7.84) A: B:

‘Right after Chairman Mao died, they started taking pictures of Committee members off the walls.’ ‘Who did they destroy more pictures of, Chairman Mao or Jiang Qing?’

It has also been long known that these constructions are strongly affected by the definiteness of the head noun, as the examples in (7.85) show. Who did you read a book about? (7.85) a. b. ?*Who did you read the book about? c. *Who did you read the green book about? d. *Who did you read Pat’s book about?

The only difference among these four sentences is the definiteness and the restrictive modification of the head noun; the basic syntactic structure of the four is the same. Hence whatever rules out (b)–(d) cannot be syntactic in nature. The NP headed by book is increasingly presupposed in (b)–(d), with concomitantly increasing difficulty of extracting out of it. Relative clauses in some languages can be affected by the definiteness and semantic content of the head noun. In Danish, for example, it is possible to form a question out of a relative clause if the main clause is relatively empty semantically and the head noun is non-specific or generic, according to Erteschik-Shir (1973) and Erteschik-Shir and Lappin (1979); they show that the Danish equivalent of what are there many who like? is perfectly acceptable. Jensen (2000) shows that even apparently definite restrictive relatives can permit extraction, as exemplified in (7.86). (7.86)

Deti kender jeg den mand som har købt that know 1sg the man CLM has bought ‘That I know the man who bought.’

i.

Such a construction is acceptable only if it immediately follows an utterance introducing the referent of det ‘that’, as in (7.87).

289

290


(7.87)

Speaker 1: Jeg tror det faldefœrdige funkishus er blevet solgt. 1sg think the tumble.down cubisthouse is been sold ‘I think the tumble-down cubist house has been sold.’ Speaker 2: Deti kender jeg den mand som har købt i . ‘That [= cubist house] I know the man who bought.’

Outside of this context (7.86) is unacceptable; hence language-specific contextual constraints serve to override the basic principle in (7.68). These facts show that restrictions on WH-question formation and related constructions cannot be treated as a purely structural phenomenon but rather must be seen as involving the interaction of syntactic structure, pragmatic functions and lexical semantics. Indeed, the fact that the same syntactic configuration may permit WH-question formation in one context or with one main verb but not allow it in a different context or with a different main verb is strong evidence that the restrictions are not purely syntactic in nature. It seems to be the case that (7.68) is not operative in the grammar of some languages at all. Japanese appears to be such a language. Kuno (1973) and others have claimed that subjacency restrictions are not found in Japanese, because it is possible to form WH-questions, relative clauses and topicalizations out of the structures in Table 7.9. This is an oversimplification, however. As Shimojo (2002) has shown, not all WH-questions, relative clauses and topicalizations out of the structures in Table 7.9 are grammatical, and the factors that condition the acceptability of these constructions are primarily pragmatic and lexical semantic. In Japanese they seem to impose restrictions on interpretations, rather than overriding (7.68) as in English and other languages. An important example of a structure which meets the condition in (7.68) but which is nevertheless incompatible with question formation, etc., is illustrated in (7.88). (7.88) a. b. c. d.

Scully said (that) Mulder interviewed the suspect. Who did Scully say interviewed the suspect? *Who did Scully say that interviewed the suspect? Who did Scully say (that) Mulder interviewed?

The basic sentence, given in (a), meets the condition in (7.68), and, as expected, WH-question formation is possible out of it, as in (b) and (d). The embedded clause is in the potential focus domain. What is surprising is the ungrammaticality of (c); the only difference between (b) and (c) is the presence of the complementizer that in (c). The presence or absence of that has no effect on questions in which the WHword is related to a post-verbal position; it is relevant only when the WH-word is interpreted as the ‘subject’ of the clause. This phenomenon is not universal, by any means. In Lakhota, for example, the presence or absence of an overt complementizer has no effect on WH-question formation in object complements (Williamson 1984; Van Valin 1993b).


The nature of this constraint concerns the distribution of narrow focus in clauses. A WH-word is narrow focus, and in a language in which WH-words occur in situ, the position in the clause where the WH-word occurs must be a possible position for narrow focus. In a language like English in which the WH-word occurs in the precore slot, the corresponding requirement is that the position the WH-word is interpreted as filling in the clause must be a possible position for narrow focus. In section 3.2 the distinction between marked and unmarked narrow focus was introduced, with narrow focus on the final core XP being the least marked and narrow focus on the ‘subject’ being the most marked. There is another parameter of markedness as well, namely main vs subordinate clause. All languages allow narrow focus in main clauses but some prohibit it in tensed embedded clauses, e.g. Polish (Eschenberg 1999). Hence main clauses are the unmarked location for narrow focus, embedded clauses the marked location. These two markedness parameters interact, as Table 7.11 shows. Thus, in terms of these two markedness parameters, the least marked type of narrow focus is narrow focus on the ‘object’ (i.e. final XP in the core) in the main clause, while the maximally marked type is narrow focus on the ‘subject’ in an embedded clause. Table 7.11 Markedness of narrow focus

‘Object’ in main cl. ‘Subject’ in main cl. ‘Object’ in emb. cl. ‘Subject’ in emb. cl.

Position of narrow focus

Location of actual focus domain

− + − +

− − + +

In (7.82) it was shown that the ‘object’ position in a that-clause is a possible focus position; what about core-initial position? The same test as in (7.82) can be used to find out. (7.89) a.

b.

Speaker 1: Scully said Mulder talked to the detective. Speaker 2: No, Skinner. (= ‘Skinner said . . .’, = ‘Scully said Skinner talked . . .’, = ‘. . . Mulder talked to Skinner’) Speaker 1: Scully said that Mulder talked to the detective. Speaker 2: No, Skinner. (= ‘Skinner said . . .’, = ‘. . . Mulder talked to Skinner’, but ?? = ‘Scully said that Skinner talked . . .’)

In (a), it seems relatively easy to interpret Skinner as replacing any of the three NPs in speaker 1’s utterance. In (b), on the other hand, while it is easy to construe Skinner as replacing Scully or the detective in speaker 1’s utterance, it is more difficult than in (a) to interpret it as replacing Mulder. This means that it is more difficult to interpret Mulder as the focus of speaker 1’s utterance in (b) than in (a), and this correlates with the presence or absence of that. It is easiest to

291

292


interpret Mulder as the focus in (b) if that is destressed and Mulder is stressed. This seems to parallel the fact that many native speakers find (7.88c) more acceptable if that is destressed and pronounced [ðə t] rather than [ðæt]. It appears, then, that the occurrence of an overt complementizer blocks marked narrow focus on the preverbal privileged syntactic argument position, and if narrow focus is not possible in a position in a clause, then it is not possible to form a WH-question with the WH-word interpreted as having the function associated with that position. This is not an issue of the potential focus domain; that the potential focus domain extends into the embedded clause is shown by the grammaticality of the WH-question in (7.88d). Thus it appears that while unmarked narrow focus is possible in the embedded clause in (7.88d) when a complementizer is present, marked narrow focus on the ‘subject’ is not. Note that when the complementizer is absent, the ‘subject’ of the embedded clause occurs in effect in a position equivalent to the final position in the matrix core, a position which is the most unmarked focus position, and in this case marked narrow focus on the ‘subject’ of the embedded clause is indeed possible, as the grammaticality of (b) shows. This is in effect a kind of erasing of the boundary between the main and embedded clause focus structures. This account predicts that if the main clause contains a peripheralCORE adjunct, the acceptability of an embedded ‘subject’ question will be degraded, because the embedded ‘subject’ will clearly not be in the main clause unmarked focus position, and this seems to be the case, as (7.90) shows. (7.90) a. Who did Kim say talked to Dana? b. *Who did Kim say yesterday talked to Dana?

Thus, the sharper the distinction between main and embedded clause, the more difficult it is to get narrow focus on the embedded clause ‘subject’. If one were to try to form a topic construction with (7.88d), the result is predictably ungrammatical, *Mulder Scully said that interviewed the suspect. Because this sentence does not pass the negation test in (7.89b), it fails to meet Reinhart’s pragmatic aboutness criterion and therefore fails to meet the condition on topic constructions in (7.81). The issues raised above in the discussion of pragmatic aboutness apply to (7.88c) as well, and the failure to meet Reinhart’s criterion has the same consequences. Thus, the occurrence of an overt complementizer renders the ‘subject’ of the embedded clause pragmatically and syntactically inert with respect to ‘extraction’ constructions. It has been shown in this section that the much discussed and theoretically very important restrictions on WH-question formation and related constructions are the result primarily of the complex interaction of syntactic structure and focus structure; they are neither purely syntactic, nor are they purely pragmatic. As is well known, languages vary with respect to the restrictions placed on question formation, and the approach presented here makes it possible to identify the parameters along which languages will vary. There are two major ones: (1) how ‘deep’ into the sentence does (7.68) apply?, and (2) how much can lexical semantic


and other factors override (7.68)? As mentioned in chapter 6, some languages restrict the potential focus domain to matrix clauses only, and accordingly no extraction out of any kind of embedded clauses is possible. Moreover, lexical semantic and pragmatic factors can interact with (7.68) to lead to variation in acceptability of question formation, etc., within a single language and across languages.

293

References

RRG web site: http://linguistics.buffalo.edu/research/rrg.html. Abdoulaye, M. L. 1992. Aspects of Hausa morphosyntax in Role and Reference Grammar. Ph.D. dissertation, University at Buffalo. (Available on RRG web site.) Aissen, Judith. 1987. Tzotzil clause structure. Dordrecht: Reidel. 1992. Topic and focus in Mayan. Language 68:43–80. Anderson, Steven R., and Paul Kiparsky, eds. 1973. A festschrift for Moris Halle. New York: Holt, Rinehart & Winston. Andrews, Avery. 1982. The representation of case in modern Icelandic. In Bresnan, ed., 427– 503. 1985. The major functions of the noun phrase. In Shopen, ed., 62–154. Aoun, Joseph, and Yen-hui Andrey Li. 1993. The syntax of scope. Cambridge, MA: MIT Press. Ariel, Mira. 1990. Accessing noun-phrase antecedents. London and New York: Routledge. Asher, Nicholas. 1999. Discourse and the focus/background distinction. In P. Bosch and R. van der Sandt, eds., Focus: linguistic, cognitive and computational perspectives, 247–67. Cambridge: Cambridge University Press. Baker, Mark. 1988. Theta theory and the syntax of applicatives in Chichewa. Natural Language and Linguistic Theory 6:353–90. Bearth, Thomas. 1969. Phrase et discours en Toura. Cahiers Ferdinand de Saussure 25:29– 45. 1992. Constituent structure, natural focus hierarchy and focus types in Toura. Folia Linguistica 26:75–94. Belloro, Valeria. 2004. A Role and Reference Grammar account of third-person clitic clusters in Spanish. Unpublished MA thesis, University of Buffalo. (Available on RRG web site.) Bentley, Delia. 2004. Ne-cliticisation and split intransitivity. Journal of Linguistics 40:219–62. Bhat, D. N. S. 1991. Grammatical relations: the evidence against their necessity and universality. London and New York: Routledge. Bickel, Balthasar. 1993. Belhare subordination and the theory of topic. In K. Ebert, ed., Studies in clause linkage (Papers from the First Köln–Zürich Workshop), 23–55. Arbeiten des Seminars für Allgemeine Sprachwissenschaft, no. 12. Zurich: University of Zürich. 2002. Belhare. In R. LaPolla and G. Thurgood, eds., The Sino-Tibetan languages, 546–70. London: Curzon Press. 2003a. Referential density in discourse and syntactic typology. Language 79:708–36. 2003b. Clause linkage typology. Lecture series delivered at the 2003 International Role and Reference Grammar Conference, UNESP, São Jose do Rio Preto, Brazil. Blake, B. J. 1979. A Kalkatungu grammar, Pacific Linguistics B:57. Canberra: Pacific Linguistics. Bloomfield, L. 1933. Language. New York: Holt. Borkin, Ann. 1984. Problems in form and function. Norwood, NJ: Ablex.

294

References

Branigan, Holly, and Mercè Prat-Sala. 2000. A cross-linguistic perspective on discourse context and syntactic processing in language production. In DeVincenzi and Lombardo, eds., 205–26. Bresnan, Joan. 1982a. Control and complementation. In Bresnan, ed., 282–390. ed. 1982b. The mental representation of grammatical relations. Cambridge, MA: MIT Press. Bresnan, Joan, and Jonni M. Kanerva. 1989. Locative inversion in Chichewa: ˆ A case study of factorization in grammar. Linguistic Inquiry 20.1:1–50. Butler, Christopher. 2003. Structure and function: a guide to three major structural-functional theories (2 vols.). Amsterdam/Philadelphia: John Benjamins. 2004. Notes towards an incremental implementation of the Role and Reference Grammar semantics-to-syntax mapping rules for English. To appear in M. Hannay and G. Steen (eds.), The English clause: usage and structure. Amsterdam: John Benjamins. Butler, James, and Charles Peck. 1980. The use of passive, antipassive, and absolutive verbs in Tzutujil of San Pedro la Laguna. Journal of Mayan Linguistics 2.1:40–52. Bybee, Joan. 1985. Morphology. Amsterdam and Philadelphia: John Benjamins. Cain, Bruce, and James Gair. 2000. Dhivehi (Maldivian), Languages of the World / Materials 63. Munich: Lincom Europa. Carlson, Greg N. 1977. Reference to kinds in English. Ph.D. dissertation, University of Massachusetts. Carnie, Andrew. 2002. Syntax: a generative introduction. Cambridge, MA: Blackwell. Cattell, Ray. 1984. Composite predicates in English. Orlando, FL: Academic Press. Centineo, Giulia. 1986. A lexical theory of auxiliary selection in Italian. In Davis Working Papers in Linguistics 1:1–35. Published in Probus 8:223–71, 1996. Chafe, Wallace. 1987. Cognitive constraints on information flow. In R. Tomlin, ed., Coherence and grounding in discourse, 21–51. Amsterdam: John Benjamins. Chomsky, Noam. 1970. Remarks on nominalization. In Roderick Jacobs & Peter Rosenbaum, eds., Readings in English transformational grammar, 184–221. Waltham, MA: Ginn & Co. 1973. Conditions on transformations. In Anderson and Kiparsky, eds., 232–86. 1981. Lectures on government and binding. Dordrecht: Foris. Clark, Robin. 1985. The syntactic nature of logical form: evidence from Toba Batak. Linguistic Inquiry 16:663–9. Cole, Peter, ed. 1981. Radical pragmatics. New York: Academic Press. Comrie, Bernard. 1976. The syntax of causative constructions: cross-language similarities and divergences. In M. Shibatani, ed., The grammar of causative constructions, 261–312. (Syntax and semantics, 6). New York: Academic Press. 1988. Conjunction reduction in pro-drop languages: some Slavic evidence. In M. Bierwisch et al., eds., Syntax, Semantic und Lexikon, 83–7. Studia Grammatica, 29. Berlin: Akademie-Verlag. Comrie, Bernard, and Sandra Thompson. 1985. Complex nominalizations. In Timothy Shopen, ed., Language typology and linguistic description, vol. III: Grammatical categories and the lexicon, 349–98. Cambridge: Cambridge University Press. Craig, Colette. 1977. The structure of Jacaltec. Austin: Texas Press. Cutrer, Michelle. 1987. Theories of obligatory control. In Davis Working Papers in Linguistics 2:6–37. Davis: University of California. 1993. Semantic and syntactic factors in control. In Van Valin, ed., 167–95. Dahm-Draksic, Tracy. 1997. A Role and Reference Grammar analysis of case marking in Croatian. University at Buffalo MA Project. (Available on RRG web site.) Datz, Margaret. 1980. Jacaltec syntactic structures and the demands of discourse. Ph.D. dissertation, University of Colorado. DeClerck, R. 1988. Studies on copular sentences, clefts and pseudoclefts. Dordrecht: Foris.

295

296

References

Demuth, Katherine. 1989. Maturation and the acquisition of the Sesotho passive. Language 65:56–80. 1990. Subject, topic and Sesotho passive. Journal of Child Language 17:67–84. Derbyshire, Desmond, and Geoffrey Pullum, eds. 1986. Handbook of Amazonian languages, vol. I. Berlin: Mouton de Gruyter. 1989. Handbook of Amazonian languages, vol. II. Berlin: Mouton de Gruyter. De Vincenzi, Marica, and Vincenzo Lombardo, eds. 2000. Cross-linguistic perspectives on language processing. Dordrecht: Kluwer. Dik, Simon C. 1989. The theory of Functional Grammar, part 1. Dordrecht: Foris. Dixon, R. M. W. 1972. The Dyirbal language of north Queensland. Cambridge: Cambridge University Press. 1994. Ergativity. Cambridge: Cambridge University Press. Dowty, David. 1979. Word meaning and Montague Grammar. Dordrecht: Reidel. Dryer, Matthew S. 1986. Primary objects, secondary objects, and antidative. Language 62:808– 45. 1992. The Greenbergian word order correlations. Language 68.1:81–138. 1996. Focus, pragmatic presupposition and activated propositions. Journal of Pragmatics 26:475–523. 1997. Are grammatical relations universal? In J. Bybee, J. Haiman and S. Thompson, eds., Essays on language function and language type, 115–43. Amsterdam: John Benjamins. Durie, Mark. 1985. A grammar of Acehnese. Dordrecht: Foris. 1987. Grammatical relations in Acehnese. Studies in Language 11:365–99. 1988. The so-called passive of Acehnese. Language 64:104–13. Erguvanlı, Eser. 1984. The function of word order in Turkish grammar, University of California Publications in Linguistics, 106. Berkeley and Los Angeles: University of California Press. Erteschik-Shir, Nomi. 1973. On the nature of island constraints. Unpublished Ph.D. dissertation, MIT. 1997. The dynamics of focus structure. Cambridge: Cambridge University Press. Erteschik-Shir, Nomi, and Shalom Lappin. 1979. Dominance and the functional explanation of island phenomena. Theoretical Linguistics 6:41–85. 1983. Under stress: a functional explanation of sentence stress. Journal of Linguistics 19:419–53. Eschenberg, Ardis. 1999. Focus structure in Polish. Unpublished MA Project, University at Buffalo. (Available on RRG web site.) Everett, Daniel L. 1986. Pirahã. In Derbyshire and Pullum, eds. (1986), 200–325. 2002. Asymmetrical clause linkage in Wari’. Unpublished ms., University of Manchester. (Available on RRG web site.) Everett, Daniel L., and Barbara Kern. 1998. Wari’: the Pacaas-Novos language of western Brazil. London: Routledge. Faltz, Leonard. 1978. On indirect objects in universal grammar. CLS 14:76–87. Félix Armendáriz, Rolando. 2000. Las relaciones gramaticales en yaqui: un análisis en el marco de la Gramática de Rol y Referencia. MA thesis, University of Sonora. Fillmore, Charles. 1968. The case for case. In E. Bach and R. Harms, eds., Universals in linguistic theory, 1–88. New York: Holt, Rinehart & Winston. Foley, William A., and Robert D. Van Valin, Jr. 1984. Functional syntax and universal grammar. Cambridge: Cambridge University Press. 1985. ‘Information packaging in the clause’. In Shopen, ed., 282–364. Frajzynger, Zygmunt. 1995. A functional theory of matrix coding. Paper presented at the Conference on Functional Approaches to Grammar, Albuquerque, NM.

References

Franklin, Karl. 1971. A grammar of Kewa, New Guinea, Pacific Linguistics, C-16. Canberra: Australian National University. Fretheim, Thorstein, and Jeanette Gundel, eds. 1996. Reference and referent accessibility. Amsterdam/Philadelphia: John Benjamins. Fried, Robert. 2003. Iconicity: some thoughts on peripheral subordination and complement clauses in English. Unpublished ms., University at Buffalo. Givón, Talmy. 1980. The binding hierarchy and the typology of complements. Studies in Language 4:333–77. 1981. Typology and functional domains. Studies in Language 5.2:163–93. 1983. Topic continuity in discourse: an introduction. In T. Givón, ed., Topic continuity in discourse: a quantitative cross-language study, 1–42. Amsterdam and Philadelphia: John Benjamins. Goldberg, Adele. 1995. Constructions: a Construction Grammar approach to argument structure. Chicago: University of Chicago Press. González Orta, Marta Mar´ıa. 2002. Lexical templates and syntactic variation: the syntax– semantics interface of the Old English speech verb secgan. In Mairal and Pérez Quintero, eds., 281–302. Grice, H. Paul. 1975. Logic and conversation. In P. Cole and J. Morgan, eds., Speech acts, 41–58. Syntax and semantics, 3. New York: Academic Press. Guerrero Valenzuela, Lilián G. 2004. The syntax–semantics interface in Yaqui complex sentences: a Role and Reference Grammar analysis. Unpublished Ph.D. dissertation, University at Buffalo. (Available on the RRG web site.) Guerrero Valenzuela, Lilián, and Robert D. Van Valin, Jr. 2004. Yaqui and the analysis of primary object languages. International Journal of American Linguistics 70:290–319. Gundel, Jeanette K. 1988. Universals of topic–comment structure. In M. Hammond, E. Moravcsik and J. Wirth, eds., Studies in linguistic typology, 209–39. Amsterdam and Philadelphia: John Benjamins. Gundel, Jeanette K., Nancy Hedberg, and Ron Zacharski. 1993. Cognitive status and the form of referring expressions in discourse. Language 69.2:274–307. Hale, Kenneth. 1973. Person marking in Warlbiri. In Anderson and Kiparsky, eds., 308–44. Halliday, Michael A. K. 1967. Notes on transitivity and theme in English. Journal of Linguistics 3:37–81, 199–244. 1994. An introduction to Functional Grammar (2nd edn). London: Edward Arnold. Han, Jeonghan. 1999. On grammatical coding of information structure in Korean: a Role and Reference Grammar account. Unpublished Ph.D. dissertation, University at Buffalo. (Available on RRG web site.) Hansell, Mark. 1993. Serial verbs and complement constructions in Mandarin: a clause linkage analysis. In Van Valin, ed., 197–233. Hasegawa, Yoko. 1996. A study of Japanese clause linkage: the connective TE in Japanese. Stanford: CSLI. Haspelmath, Martin. 1993. More on the typology of the inchoative/causative verb alternation. In B. Comrie and M. Polinsky, eds., Causatives and transitivity, 87–120. Amsterdam: John Benjamins. 1999. Explaining article–possessor complementarity: economic motivation in noun phrase syntax. Language 75:227–43. Haumann, Dagmar. 1997. The syntax of subordination. Tübingen: Max Niemeyer Verlag. Heath, Jeffrey. 1975. Some functional relationships in grammar. Language 51:89–104. 1977. Choctaw cases. BLS 3:204–13. Heusinger, Klaus von. 1999. Intonation and information structure. Habilitationschrift, University of Konstanz.

297

298

References

Heydel, Maren, and Wayne Murray. 2000. Conceptual effects in sentence priming: a crosslinguistic perspective. In De Vincenzi and Lombardo, eds., 227–54. Hoijer, Harry. 1949. Tonkawa syntactic suffixes and anaphoric particles. Southwestern Journal of Anthropology 5:37–55. Holisky, Dee A. 1979. On lexical aspect and verb classes in Georgian. In CLS Parasession on the Elements / Non-Slavic Languages of the USSR, 390–401. Chicago: Chicago Linguistic Society. 1981a. Aspect and Georgian medial verbs. New York: Caravan. 1981b. Aspect theory and Georgian aspect. In Philip Tedeschi and Annie Zaenen, eds., Tense and Aspect, 127–44. Syntax and semantics, 14. New York: Academic Press. 1987. The case of the intransitive subject in Tsova-Tush (Batsbi). Lingua 71:103–32. Hooper, Joan B., and Sandra A. Thompson. 1973. On the applicability of root transformations. Linguistic Inquiry 4:465–98. Huang, C.-T. James. 1982. Logical relations in Chinese and the theory of grammar. MIT dissertation. Hyman, Larry, and Karl Zimmer. 1976. Embedded topic in French. In Li, ed., 189–211. Ikegami, Y. 1985. ‘Activity’ – ‘Accomplishment’ – ‘Achievement’: a language that can’t say ‘I burned it but it didn’t burn’ and one that can. In A. Makkai and A. Melby, eds., Linguistics and philosophy: essays in honor of Rulon S. Wells, 265–304. Amsterdam: John Benjamins. Imai, Shingo. 1998. Logical structures and case marking in Japanese. Unpublished MA project, University at Buffalo. (Available on RRG web site.) Ioup, G. 1975. Some universals for quantifier scope. In J. Kimball, ed., Syntax and semantics, 4, 37–58. New York: Academic Press. Jackendoff, Ray S. 1972. Semantic interpretation in generative grammar. Cambridge, MA: MIT Press. 1976. Toward an explanatory semantic representation. Linguistic Inquiry 7:89–150. 1977. X-bar syntax. Cambridge, MA: MIT Press. 1990. Semantic structures. Cambridge, MA: MIT Press. 1992. Madame Tussaud meets the Binding Theory. Natural Language and Linguistic Theory 10:1–31. 1996. The proper treatment of measuring out, telicity, and possibly even quantification in English. Natural Language and Linguistic Theory 14:305–54. Jensen, Anne. 2000. Sentence intertwining in Danish – a challenge to the RRG account? Paper presented at the Conference on New Theoretical Perspectives on Syntax and Semantics in Cognitive Science, Dubrovnik, Croatia. (Available on RRG web site.) Johnson, Mark. 1987. A new approach to clause structure in Role and Reference Grammar. In Davis Working Papers in Linguistics 2:55–9. Davis: University of California. Jolly, Julia. 1991. Prepositional analysis within the framework of Role and Reference Grammar. New York: Peter Lang. 1993. Preposition assignment in English. In Van Valin, ed., 275–310. Jones, W., and P. Jones. 1991. Barasano syntax, Studies in the Languages of Colombia, 2. Dallas: SIL. Kamp, Hans, and Uwe Reyle. 1993. From discourse to logic. Hingham, MA: Kluwer. Keenan, Edward L. 1976. Remarkable subjects in Malagasy. In Li, ed., 247–301. Kim, Alan H. O. 1988. Preverbal focusing and type XXIII languages. In M. Hammond, E. Moravcsik and J. Wirth, eds., Studies in syntactic typology, 147–72. Amsterdam and Philadelphia: John Benjamins. Kimenyi, Alexandre. 1980. A relational grammar of Kinyarwanda, UCPL 91. Berkeley: University of California Press. Klamer, Marian. 1994. Kambera: a language of eastern Indonesia, HIL dissertations, 11. Amsterdam: Holland Institute of Generative Linguistics.

References

Kluender, Robert. 1992. Deriving island constraints from principles of predication. In Helen Goodluck and Michael Rochmont, eds., Island constraints, 223–58. Dordrecht: Kluwer. Kockelman, Paul. 2003. The interclausal relations hierachy in Q’eqchi’ Maya. International Journal of American Linguistics. 69:25–48. Kuno, Susumu. 1972a. Functional sentence perspective: a case study from Japanese and English. Linguistic Inquiry 3:269–320. 1972b. Pronominalization, reflexivization, and direct discourse. Linguistic Inquiry 3:161–96. 1973. The structure of the Japanese language. Cambridge, MA: MIT Press. 1975. Three perspectives in the functional approach to syntax. In Papers from the Parasession on Functionalism, 276–336. Chicago: Chicago Linguistic Society. 1987. Functional syntax: anaphora, discourse, and empathy. Chicago: University of Chicago Press. 1991. Remarks on quantifier scope. In H. Nakajima, ed., Current English linguistics in Japan, 261–87. Berlin and New York: Mouton de Gruyter. Kuno, S., K. Takami and Y. Wu. 1999. Quantifier scope in English, Chinese and Japanese. Language 75:63–111. Kwee Tjoe Liong. 2002. More-than-complex sentences – modifiers, layering, and recursion, in Functional Grammar and in Role and Reference Grammar. Paper presented at 2002 RRG Conference, Universidad de La Rioja, Spain. Lambrecht, Knud. 1986. Topic, focus and the grammar of spoken French. Unpublished Ph.D. dissertation, University of California, Berkeley. 1987. Sentence focus, information structure, and the thetic-categorial distinction. BLS 13:366–82. 1994. Information structure and sentence form. Cambridge: Cambridge University Press. 2000. When subjects behave like objects: a markedness analysis of sentence focus constructions across languages. Studies in Language 24:611–82. Langacker, Ronald. 1995. Raising and transparency. Language 71:1–62. LaPolla, Randy. 1990. Grammatical relations in Chinese. Unpublished Ph.D. dissertation, University of California, Berkeley. Levinson, Stephen C. 1987. Pragmatics and the grammar of anaphora: a partial pragmatic reduction of binding and control phenomena. Journal of Linguistics 23:379–434. Li, Charles, ed. 1976. Subject and topic. New York: Academic Press. Li, Charles, and Rainer Lang. 1979. The syntactic irrelevance of an ergative case in Enga and other Papuan languages. In F. Plank, ed., Ergativity, 307–24. London: Academic Press. Lyons, John. 1968. Introduction to Theoretical Linguistics. Cambridge: Cambridge University Press. Mairal, R. and P. Faber. 2002. Functional Grammar and lexical templates. In Mairal and Pérez Quintero, eds., 39–94. Mairal, R. and M. J. Pérez Quintero, eds. 2002. New perspectives on argument structure in Functional Grammar. Berlin / New York: Mouton de Gruyter. Maling, Joan. 1986. Clause-bounded reflexives in Modern Icelandic. In L. Hellan and K. Christensen, eds., Topics in Scandinavian syntax, 53–64. Dordrecht: Reidel. Matasović, Ranko, 2001. Adjective phrases. Unpublished ms., University of Zagreb. (Available on RRG web site.) Mathesius, Vilém. 1928. On linguistic characterology with illustrations from modern English. Actes du Premier Congrès international de linguistes a` la Haye, 56–63. [Reprinted in Vachek (1964:59–67).] 1929. Functional linguistics. In Vachek (1983:121–42). Matthiessen, Christian, and Sandra Thompson. 1988. The structure of discourse and ‘subordination’. In J. Haiman and S. Thompson, eds., Clause combining in grammar and discourse, 275–330. Amsterdam and Philadelphia: John Benjamins.

299

300

References

McConnell-Ginet, Sally. 1982. Adverbs and logical form. Language 58:144–84. McLendon, Sally. 1978. Ergativity, case, and transitivity in Eastern Pomo. International Journal of American Linguistics 44:1–9. Melinger, Alissa. 1996. Quantifier scope in Italian: a syntactic comparison. In Eve Ng and Corinne Grimm, eds., Proceedings from the Second Buffalo-Toronto Student Conference in Linguistics, 160–78. Buffalo: University at Buffalo Linguistics Dept. Michaelis, Laura. 1993. On deviant case marking in Latin. In Van Valin, ed., 311–73. Mohanan, K. P. 1982. Grammatical relations and clause structure in Malayalam. In Bresnan, ed., 504–89. Narasimhan, B. 1998. A lexical semantic explanation for ‘quirky’ case marking in Hindi. Studia Linguistica 52:48–76. Nichols, Johanna. 1986. Head-marking and dependent-marking grammar. Language 62:56– 119. Noonan, Michael. 1992. A grammar of Lango. Berlin and New York: Mouton de Gruyter. Noonan, Michael, and Edith Bavin-Woock. 1978. The passive analog in Lango. BLS 4:28–39. Nunes, Mary L. 1993. Argument linking in English derived nominals. In Van Valin, ed., 375– 432. Ohori, Toshio. 1996. Case markers and clause linkage: toward a semantic typology. In E. Casad, ed., Cognitive linguistics in the redwoods, 693–712. Berlin: Mouton de Gruyter. 2001. Some thoughts on a new systematization of interclausal semantic relations. Paper presented at 2001 Role and Reference Grammar Conference, University of California, Santa Barbara. (Available on RRG web site.) Olson, Michael L. 1978. Switch-reference in Barai. BLS 4:140–57. 1981. Barai clause junctures: toward a functional theory of interclausal relations. Unpublished Ph.D. dissertation, Australian National University. Osborne, C. R. 1974. The Tiwi language. Canberra: Australian Institute of Aboriginal Studies. Paris, Luis. 2003. Grammatical encoding of event relations: gerund phrases in Spanish. Unpublished Ph.D. dissertation, University at Buffalo. Park, Ki-seong. 1993. Korean causatives in Role and Reference Grammar. MA thesis, SUNY at Buffalo. (Available on RRG web site.) Pavey, Emma. 2001. Information structure in Acadian French. Unpublished MA thesis, University of North Dakota. (Available on RRG web site.) 2004. The English IT-cleft construction: a Role and Reference Grammar analysis. Unpublished D.Phil. dissertation, University of Sussex. (Available on RRG web site.) Payne, Doris L., and Thomas E. Payne. 1989. Yagua. In Derbyshire and Pullum, eds. (1989), 249–474. Pollard, Carl, and Ivan A. Sag. 1992. Anaphors in English and the scope of binding theory. Linguistic Inquiry 23.2:261–304. Prince, Ellen. 1981a. Topicalization, focus movement and Yiddish movement: a pragmatic differentiation. BLS 7:249–64. 1981b. Toward a taxonomy of given–new information. In Cole, ed., 223–56. Pustejovsky, James J. 1991. The generative lexicon. Computational Linguistics 17:409–41. 1995. The generative lexicon. Cambridge, MA: MIT Press. Rappaport Hovav, Malka, and Beth Levin. 1998. Building verb meanings. In M. Butt and W. Geuder, eds., The projection of arguments: lexical and compositional factors, 97–134. Stanford: CSLI. Reinhart, Tanya. 1981. Pragmatics and linguistics: an analysis of sentence topics. Philosophica 27:53–94. Rijkhoff, Jan. 1990. Explaining word order in the noun phrase. Linguistics 28:5–42. 2002. The noun phrase: a typological study of its form and structure. Oxford: Oxford University Press.

References

Roberts, John. 1987. Amele. London: Croom Helm. 1988. Amele switch-reference and the theory of grammar. Linguistic Inquiry 19.45–64. 2002. Distributives in Role and Reference Grammar. Unpublished ms. Roberts, Linda. 1995. Pivots, voice and macroroles: from Germanic to universal grammar. Australian Journal of Linguistics 15:157–214. Ross, John Robert. 1967. Constraints on variables in syntax. Unpublished Ph.D. dissertation, MIT. 1972. Act. In D. Davidson and G. Harman, eds., Semantics of natural language, 70–126. Dordrecht: Reidel. Sapir, Edward. 1921. Language. New York: Henry Holt. Schwartz, Linda. 1993. On the syntactic and semantic alignment of attributive and identificational constructions. In Van Valin, ed., 433–63. Selkirk, Elizabeth. 1984. Phonology and syntax. Cambridge, MA: MIT Press. 1995. Sentence prosody: intonation, stress and phrasing. In J. Goldsmith, ed., The handbook of phonological theory, 550–69. Oxford: Blackwells. Sgall, Petr, Eva Hajiˇcová and Jarmila Panevová. 1986. The meaning of the sentence in its semantic and pragmatic aspects, ed. Jacob L. Mey. Dordrecht and Boston: Reidel. Shimojo, Mitsuaki. 1995. Focus structure and morphosyntax in Japanese: WA and GA, and word order flexibility. Ph.D. dissertation, University at Buffalo. (Available on RRG web site.) 2002. Functional theories of island phenomena: the case of Japanese. Studies in Language 26:67–124. Shopen, Timothy, ed. 1985. Language typology and syntactic description, vol. I: Clause structure. Cambridge: Cambridge University Press. Silverstein, Michael. 1976. Hierarchy of features and ergativity. In R. M. W. Dixon, ed., Grammatical categories in Australian languages, 112–71. Canberra: Australian Institute of Aboriginal Studies. 1981. Case marking and the nature of language. Australian Journal of Linguistics 1:227–46. 1993. Of nominatives and datives: Universal grammar from the bottom up. In Van Valin, ed., 465–98. Smith, Carlotta. 1997. The parameter of aspect (2nd edn). Dordrecht: Reidel. Soames, Scott. 1982. How presuppositions are inherited: a solution to the projection problem. Linguistic Inquiry 13:483–545. Stahlke, H. 1970. Serial verbs. Studies in African Linguistics 1.60–99. Sugamoto, Nobuko. 1984. Reflexives in Toba Batak. In P. Schachter, ed., Studies in the structure of Toba Batak, 150–71. UCLA Occasional Papers in Linguistics, 5. Los Angeles: UCLA Department of Linguistics. Suzuki, N. 1991. Focus and conditions on representations. Lingua 83:183–248. Swadesh, Morris. 1939. Nootka internal syntax. International Journal of American Linguistics 9:77–102. Takami, Ken-ichi. 1988. Preposition stranding: arguments against syntactic analyses and an alternative functional explanation. Lingua 76:299–335. Tallerman, Maggie. 1998. Understanding syntax. London: Edward Arnold. Talmy, Leonard. 2000. Toward a cognitive semantics. Cambridge, MA: MIT Press. Thráinsson, Höskuldur. 1979. On complementation in Icelandic. New York: Garland. 1991. Long-distance reflexives and the typology of NPs. In J. Koster and E. Reuland, eds., Long-distance anaphora, 49–75. Cambridge: Cambridge University Press. Thurman, Robert. 1975. Chuave medial verbs. Anthropological Linguistics 17.7:342–52. Tomcsányi, Judit. 1988. Roles y Referencia en Bribri: aspectos de la determinación funcional en las structuras sintácticas. Unpublished Ph.D. dissertation, National University of Costa Rica.

301

302

References

Toratani, Kiyoko. 1998. Lexical aspect and split intransitivity in Japanese. CLS 34:377–92. 2002. The morphosyntax and the logical structures of compound verbs in Japanese. Unpublished Ph.D. dissertation, University at Buffalo. (Available on RRG web site.) Vachek, J., ed. 1964. A Prague School reader in linguistics. Bloomington, IN: Indiana University Press. Vallduv´ı, Enric. 1992. The informational component. New York: Garland. Van Valin, Robert D., Jr. 1977. Aspects of Lakhota syntax. Unpublished Ph.D. dissertation, University of California, Berkeley. 1981. Grammatical relations in ergative languages. Studies in Language 5:361–94. 1985. Case marking and the structure of the Lakhota clause. In J. Nichols and A. Woodbury, eds., Grammar inside and outside the clause, 363–413. Cambridge: Cambridge University Press. 1987. The role of government in the grammar of head-marking languages. International Journal of American Linguistics 53:371–97. 1990. Semantic parameters of split intransitivity. Language 66:221–60. 1991. Another look at Icelandic case marking and grammatical relations. Natural Language and Linguistic Theory 9:145–94. ed. 1993a. Advances in Role and Reference Grammar. Amsterdam and Philadelphia: John Benjamins. 1993b. A synopsis of Role and Reference Grammar. In Van Valin, ed., 1–164. 1995. Toward a functionalist account of so-called extraction constraints. In Betty Devriendt, Louis Goossens and Johan van der Auwera, eds., Complex structures: a functionalist perspective, 29–60. Berlin: Mouton de Gruyter. 1998. The acquisition of WH-questions and the mechanisms of language acquisition. In Michael Tomasello, ed., The new psychology of language: cognitive and functional approaches to language structure, 221–49. Mahwah, NJ: LEA. (Available on RRG web site.) 1999a. Generalized semantic roles and the syntax–semantics interface. In F. Corblin, C. Dobrovie-Sorin and J.-M. Marandin, eds., Empirical issues in formal syntax and semantics 2, 373–89. The Hague: Thesus. (Available on RRG web site.) 1999b. A typology of the interaction of focus structure and syntax. In E. Raxilina and J. Testelec, eds., Typology and linguistic theory: from description to explanation, 511–24. Moscow: Languages of Russian Culture. (Available on RRG web site.) 2001a. The acquisition of complex sentences: a case study in the role of theory in the study of language development. CLS 36.2:511–31. 2001b. An introduction to syntax. Cambridge: Cambridge University Press. 2002. The development of subject–auxiliary inversion in English wh-questions: an alternative analysis. Journal of Child Language 29:161–75. 2003. Minimalism and explanation. In J. Moore and M. Polinsky, eds., Explanation in linguistics, 281–97. Stanford: CSLI. (Available on RRG web site.) 2004. Lexical representation, co-composition, and linking syntax and semantics. Unpublished ms., University at Buffalo. (Available on RRG web site.) 2005. On the relationship between syntactic theory and models of language processing. In A. Bornkessel and Matthias Schlesewsky, eds., Semantic role universals and argument linking: theoretical, typological and psycho-/neurolinguistic perspectives. Berlin: Mouton de Gruyter. (Available on RRG web site.) Van Valin, Robert D., Jr., and Randy J. LaPolla. 1997. Syntax: structure, meaning and function. Cambridge: Cambridge University Press. Van Valin, Robert D., Jr., and David P. Wilkins. 1993. Predicting syntactic structure from semantic representations: remember in English and Mparntwe Arrernte. In Van Valin, ed., 499–534.

References

1996. The case for ‘effector’: case roles, agents and agency revisited. In M. Shibatani and S. Thompson, eds., Grammatical constructions, 289–322. Oxford: Oxford University Press. Vendler, Zeno. 1967. Linguistics in philosophy. Ithaca: Cornell University Press. Verkuyl, Henk J. 1972. On the compositional nature of the aspects. Dordrecht: Reidel. 1993. A theory of aspectuality. Cambridge: Cambridge University Press. Walton, Charles. 1986. Sama verbal semantics: classification, derivation and inflection. Manila: Linguistic Society of the Philippines. Wasow Thomas. 1977. Transformations and the lexicon. In Peter Culicover, Thomas Wasow and Adrian Akmajian, eds., Formal syntax, 327–60. New York: Academic Press. Watanabe, K. 1995. Quantification and focus structure in Japanese. Unpublished paper, SUNY Buffalo. Watters, James K. 1986. Notes on Tepehua verbal semantics. Davis Working Papers in Linguistics 1:118–44. 1988. Topics in Tepehua grammar. Unpublished Ph.D. dissertation, University of California, Berkeley. 1993. An investigation of Turkish clause linkage. In Van Valin, ed., 535–60. Weber, David J. 1989. A grammar of Huallaga (Huanuco) Quechua, University of California Publications in Linguistics 112. Berkeley: University of California Press. Weist, R. M. 2002. The first language acquisition of tense and aspect: a review. In R. Salaberry and Y. Shirai, eds., Tense–aspect morphology in L2 acquisition, 21–78. Amsterdam: John Benjamins. Weist, R. M., A. Pawlak and J. Carapella. 2004. Syntactic–semantic interface in the acquisition of verb morphology. Journal of Child Language 31:31–60. Whaley, Lindsay J. 1993. The status of obliques in linguistic theory. Ph.D. dissertation, University at Buffalo. Wilkins, David P. 1988. Switch-reference in Mparntwe Arrernte (Aranda): form, function, and problems of identity. In Peter Austin, ed., Complex sentence constructions in Australian languages, 141–76. Amsterdam and Philadelphia: John Benjamins. 1989. Mparntwe Arrernte (Aranda): studies in the structure and semantics of grammar. Unpublished Ph.D. dissertation, Australian National University. 1991. The semantics, pragmatics and diachronic development of ‘associated motion’ in Mparntwe Arrernte. In Buffalo Papers in Linguistics 91–01:207–57. Department of Linguistics, University at Buffalo. Williamson, Janice. 1984. Studies in Lakhota grammar. Unpublished Ph.D. dissertation, University of California, San Diego. Winther-Nielsen, Nicolai. 1995. A functional discourse grammar of Joshua: a computerassisted rhetorical structure analysis. Stockholm: Almqvist & Wiksell. Yang, Byong-seon. 1994. Morphosyntactic phenomena of Korean in Role and Reference Grammar: psych-verb constructions, inflectional verb morphemes, complex sentences, and relative clauses. Ph.D. dissertation, State University of New York at Buffalo. Seoul: Hankuk Publishers. Zaenen, Annie, Joan Maling and Hölskuldur Thráinsson. 1985. Case and grammatical functions: the Icelandic passive. Natural Language and Linguistic Theory 3:441–83. Zovko, Irena. 2000. Dative shift in Croatian. Paper presented at the Conference on New Theoretical Approaches to Syntax and Semantics in Cognitive Science, Dubrovnik. 2001. Semantiˇcko-sintaktiˇcki odnosi u reˇcenicama s dvostruko prijelaznim glagolima u engleskom jeziku. MA thesis, University of Zagreb.

303

Languages index

Acehnese 66, 90–3, 94, 101, 106, 145, 242, 251 Amele 10–11, 106, 107, 165, 184–7

Hausa 32 Hebrew, Biblical 213 Hindi 108

Barai 39, 106, 188, 191, 202–4 Barasano 10 Belhare 100, 112, 127, 204 Bribri 32

Icelandic 4, 32, 35, 94, 100, 102, 108, 112, 117–20, 161, 166, 173, 251–2, 254, 256, 259, 268, 270 Indonesian 117 Ingush 107 Italian 5, 32, 70–2, 74, 75, 76–7, 83, 85, 100, 117, 174–8

Chechen 107 Chichewa ˆ 72, 121 Chinese, see Mandarin Choctaw 19, 107 Chuave 184, 187, 201, 205 Cree 107 Croatian 18, 19, 27, 32, 70–2, 77, 108, 110, 111, 114, 121, 231 Danish 289 Dhivehi 6, 179, 273 Dyirbal 1, 4, 5, 13, 16, 29, 65, 94, 100, 105, 110, 111, 114, 121, 127, 144–5, 154, 161, 241, 249 Eastern Pomo 106 Enga 107, 151 English 1, 4, 5, 7, 8, 9–10, 11, 13–16, 18, 24–7, 32–40, 42–9, 50–7, 59, 60–1, 63, 65, 66, 70, 72, 74, 75, 76–7, 81–3, 85, 89, 92, 94–5, 97, 99, 101–2, 103–4, 105, 108, 113–15, 121, 126, 130, 132, 133, 137–43, 151, 153, 155–6, 159, 161, 164–9, 180–2, 188–96, 197–200, 202, 205, 206–7, 210–11, 212–13, 214–16, 221–3, 228, 233, 238, 243–4, 245–8, 251, 252–4, 256, 259, 260, 265, 268–72, 273, 274, 278, 282–3, 284, 285–6, 288–9, 290–2 Old English 46, 119 Fore 107 French 41, 70–2, 74, 75, 77, 188, 190, 192, 234–8 Georgian 32, 33, 41, 100 German 46, 57, 64, 65–6, 100, 108, 112, 116, 117, 118, 120, 143, 166, 169, 193, 212, 216

304

Jakaltek 65, 99, 100, 105, 134, 191 Japanese 6, 8, 9, 32, 33, 43, 57, 70–2, 73, 83–4, 100, 159–61, 173, 290 Kalkatungu 97–8, 100, 101–2, 117 Kambera 97, 101 Kewa 9–10, 107, 183–4, 187, 201, 202 Kinyarwanda 74, 100, 121, 251 Korean 10, 32, 43, 72, 100 Lakhota 1, 4, 8, 26, 32, 39, 45, 46, 57, 65, 107, 123, 128, 130, 147, 156, 161, 179, 196, 209, 210, 212, 216–20, 241, 273, 274–7, 278, 282 Lango 105 Latin 18, 108, 116 Maithili 100 Malagasy 105, 260, 263 Malayalam 4 Mandarin 74, 79, 85, 166–74, 188, 191 Mparntwe Arrernte 32, 43–4, 46, 107, 201, 206, 211 Nepali 100, 112, 204 Nootka 28 Pirahã 33 Polish 18, 77, 291 Portuguese 26, 77 Q’eqchi’ 209 Quechua 9, 41, 46, 47, 74, 260

Languages index Russian 16, 18, 41, 47, 65–6, 77, 108, 109, 110, 118 Sama 32, 99, 100, 105, 117, 121, 132–5, 147, 157, 166, 169, 242 Sesotho 75, 77 Setswana 75, 77 Spanish 20, 49, 190 Swahili 26 Tagalog 1, 32, 48, 105 Tepehua 32, 104, 105, 107, 229, 230 Thai 107, 166, 173 Tibetan 100 Tiwi 10 Toba Batak 72, 77, 86–8, 163–4 Tonkawa 201 Toura 75–6, 77

Turkish 9–10, 11, 35, 72, 116, 179, 203, 205, 209, 273 Tzotzil 4, 6–7, 9, 29, 57, 65 Tzutujil 105 Ute 116 Vietnamese 107 Wari’ 200 Warlpiri 107, 108, 109, 110 Yagua 23, 41 Yaqui 124–7, 209, 236 Yatye 32, 65 Yimas 107 Yoruba 65 Zapotec 5, 107

305

Subject index

aboutness, pragmatic 285 absolutive case 108 accusative case 108, 258 accusativity, syntactic 97, 99–102 acquisition, language 2, 32, 190 activation 79, 137, 175 actor, see macroroles, semantic actor–undergoer hierarchy 61–2, 121, 123, 126, et passim adjectives 26, 48 adjuncts 19–21 adpositional phrase, layered structure of 23 adpositions 19, 21 assignment rules 113 non-predicative 21–2, 23, 181 predicative 21–3, 49, 181 adjunct 22, 49, 194 argument-adjunct 22–3, 49, 168 semantic representation of 49 adverbial clauses, see subordinate clause adverbs 19–21, 26, 36–7, 49 affectedness 114, 168 agent 53, 56–7 agreement 89, 90, 101, 108, 112 Aktionsart 31–49 accomplishment 31–41, 42–6, 54 achievement 31–40, 42, 47, 54 active accomplishment 32–40, 41, 44–5, 47 activity 31–40, 41–7, 54, 56, 63 causative types 34, 39–40, 41, 42, 57 semelfactive 33–40, 42, 47, 54 state 31–46, 48–9, 50, 54 tests for 34–41 animacy 108, 123 antipassive, see voice applicatives 115, 121–3, 152 appositives 223 argument-adjunct, see adpositions, predicative argument structure, see logical structures arguments 4, et passim semantic 8 sharing 190, 240 syntactic, see core arguments

306

aspect nominal 24 verbal 8–11, 35–6, 202 assertion, pragmatic 69, 171 be 13, 48–9 branching, direction of 10, 12 case marking 108 accusative pattern 108 ergative pattern 108 in complex sentences 258–60 causation, see Aktionsart causative construction 124, 160, 235–8, 243 causative paraphrase test 38–9 clause constraints on theory of 3, 18, 29 formal representation of, see projection grammar layered structure of 3–8, 27, 183 non-universal features of 8 universal features of 8 clause linkage asymmetrical 198, 200 symmetrical 199, 200 clause-linkage marker 205 cleft construction 72, 73 complement selection 210–11 complementation, see subordinate clause complementizer, see clause-linkage marker completeness constraint 129, 245, et passim complex sentences, see juncture; nexus conjunction reduction 95, 101, 229–31 (see also topic, topic chain) constituent projection, see projection grammar constructional schemas 131–5, 147, 156, 230 control, obligatory 241–4, 249 control construction 91, 94, 95, 101, 241–50, 268 controller, see privileged syntactic argument core 4, et passim core arguments 4, 8, 17, 18, 57, 115 direct 7, 57, 65 oblique 7, 57 cross-reference, see agreement

Subject index dative case 64, 109–11, 118–19, 145 dative shift 121, (see also macroroles, semantic, undergoer, variable linking) dependence, operator, see operators, sharing dependent-marking languages 16–19 adposition structure 23 clause structure 18 possessive NP structure 27 directionals 9–10, 11 Discourse Representation Theory 171–4, 233 double-marking languages 19 ergative case 108 ergativity, syntactic 97, 100 event quantification 11 evidentials 9–11 exceptional case marking, see matrix-coding constructions focus 68, 171 broad 69 completive 72, 76, 78, 172 contrastive 72, 76, 78, 172 narrow 69, 71, 78, 82, 172, 229, 291 marked 72, 291 unmarked 72, 233, 291 predicate 70, 82, 103, 172, 229 sentence 70–1, 178 focus domain 70 actual 75, 77, 171, 215, 275, 282, 283, 286, 289 lexical influences on 288–9 potential 75, 77, 83, 171, 214–16, 219, 275–8, 282–4, 285, 286, 288–9 focus position, unmarked 72, 86 focus-sensitive elements 216–18 focus structure 69 and word order 76, 174 projection, see projection grammar Functional Grammar 2 gerund 189, 196 grammatical relations 67, 89 non-universality of 89, 93 head-marking languages 16–19, 147, 156 adposition structure 23 clause structure 16, 17–18 possessive NP structure 17 iconicity 209 illocutionary force 9–11, 77, 202, 208, 214, 229, 284 incorporation 8, 146 infinitival complements 189–90 information unit 77, 215

instrument 58–9 instrumental case 109–11, 145 Interclausal Relations Hierarchy 208, 209, 210, 245 semantic hierarchy 208 syntactic hierarchy 208 interclausal semantic hierarchies 211 causal 211–13 necessarily shared participant 211–13 participant’s mental disposition 211–13 temporal 211–13 interclausal semantic relations 206–8, 211 causative [1] 206, 207, 208, 209, 212 causative [2] 206, 207, 208, 212 circumstances 206, 207 cognition 206, 207 concessive 207, 284 conditional 207, 284 discourse, direct 206, 207, 208, 283 discourse, indirect 206, 207, 208 jussive 206, 207, 213 manner 206, 207 means 206, 207 motion 206, 207 perception, direct 206, 207, 213 perception, indirect 206, 207, 213 phase 206, 207 position 206, 207 propositional attitude 206, 207, 208, 210 psych-action 206, 207, 210, 213, 245 purposive 206, 207, 208, 213 reason 206, 207, 284 temporal 207 sequential 207 simultaneous 207 unspecified 207 intonation 73 island constraints, see relative clauses, constraints on; topicalization, constraints on; questions, WH-questions, constraints on juncture 188–220, (see also juncture–nexus combinations) clausal 188, 225, 228–34 core 188, 225, 259 nuclear 188, 190–1, 225 sentential 192 juncture–nexus combinations 191, 197 clausal coordination 198, 202 clausal cosubordination 197, 198, 229 clausal subordination 193, (see also subordinate clause) daughter 198 peripheral (ad-clausal) 198, 282 core coordination 197, 198, 202 core cosubordination 197, 198, 202

307

308

Subject index

juncture–nexus combinations (cont.) core subordination 197, (see also subordinate clause) daughter 198, 238–9, 240 peripheral ad-core 198, 200, 217, 240, 282, 283 nuclear coordination 202 nuclear cosubordination 198, 202 nuclear subordination 196, 202 sentential coordination 197, 198 sentential subordination 192, 197, 198 left-detached position 7, 76, 192 lexical content, inherent, see animacy lexical decomposition 42–6, (see also logical structures) lexical entry 66–7, 118, 211 lexical phenomena 158, 160 vs syntactic phenomena 158 lexical rules 47, 48, 66, 158 lexicon 31, 158, 161 linking in complex sentences semantics to syntax 228, 232, 236, 245, 252, 256 algorithm 228, 279 syntax to semantics 240, 249, 253, 256 algorithm 229, 250, 254, 280 linking in simple sentences role of discourse-pragmatics in 170, 182 semantics to syntax 129, 130, 134 algorithm 137, 225 syntax to semantics 129, 131, 134, 149 algorithm 136–50, 226 logical structures 42–9, 53–4, 56, 57–60, et passim macroroles, semantic 53, 60–7, 93 actor 53, 60–7, 89, 91, 117, 137, et passim default assignment principles 62–3, 118 undergoer 53, 60–7, 89, 92, 115, 117, 137, et passim variable linking 61, 111, 114, 122, 139, 143, 152 matrix-coding constructions 93, 96, 119, 250–7, 268, 271 as non-PSA (‘raising to object’) 254–7 as PSA (‘raising to subject’) 251–4 modality 8–11, 202 mood 186, (see also illocutionary force) ne-cliticization 175–7 negation 9–10, 186 neutralization, restricted 90, 94, 97, 99, 106 patterns 97–8

nexus 183, 191, (see also juncture–nexus combinations) coordination 183, 185, 192, 203, 223 vs conjunction 187 cosubordination 187, 192, 202, 203, 204, 223 non-subordinate 190, 225, 240 subordination 183–4, 190, 205, 223 ad-subordination 194, 197 daughter 197 nominative case 108, 258 noun complement construction 221 noun phrase 23–8 discontinuous 28–9, 121, 177 juncture–nexus types coren cosubordination 221 coren subordination 221 NP coordination 221 NP cosubordination 221 NP subordination 221, 222 nuclearn cosubordination 221 nuclearn subordination 221 layered structure of 24–7 coren 24 NP 24 nucleusn 24 peripheryn 24–6 representation of 24 operators 24, 52, 177 semantic representation of 50 NP-final position 27 NP-initial position 26–7 nucleus 4, 8, 13, 28, 200 object direct 67, 115, 121 indirect 115 primary 123 primary object languages 123 secondary 123 oblique constituents core, see core arguments, oblique peripheral, see adjuncts operator projection, see projection grammar operators 8–13, 26, 48–9, 50 clausal 9 core 9 nuclear 8 ordering 11 semantic representation of 50 sharing 185, 187, 201–5 parser 129, 131 passive, see voice periphery 4, 7, 194, 196, 228 pivot, see privileged syntactic argument

Subject index possession 52 alienable 52 inalienable 52, 146 kin possession 52 possessive NP 27 semantic representation 52 possessor ‘raising’ 92, 145, 208 postcore slot 5–6, 76, 179 postpositions, see adpositions precore slot 5, 6–7, 132, 179, 193, et passim predicate 4, 8 preposition stranding 142, 155, 180 prepositions, see adpositions presentational construction 71, 104, 178 presupposition pragmatic 69, 171 privileged syntactic argument 94–6, 99, 105, et passim controller 95–6 pragmatic 105, 229 semantic 101, 102 syntactic 101: invariable 102; variable 102 pivot 95–6 pragmatic 105, 229 semantic 101 syntactic 101: invariable 102; variable 102 selection principles 100, 107, 112, 117, 166 processing, language 2, 129 projection grammar 11–13 constituent projection 12–13 focus structure projection 77, 171–3 operator projection 12–13 pronouns 52, 108 punctual 33, 34, 38, 47 qualia 50–2 quantifier hierarchy 85 quantifiers 81–8 questions WH-questions 85, 132, 134, 140, 147, 273 constraints on 275–8 core-internal focus position 179 in-situ 179, 180 yes–no questions 179, 215, 216, 218 constraints on 219 raising, see matrix-coding constructions reciprocals 164 referent, coding of 73 referential phrase 28 reflexivization 134, 147, 159 clitic 161 coreference 161–2 domain of obligatory 167, 267, 268, 269, 270 domain of possible 167 lexical 161

reflexive pronouns 52 role hierarchy condition 162–4, 267 superiority condition 166, 267, 271 relative clauses 260, 273, 285 constraints on 287–8 internally headed 260 linking 261, 263–4, 265 non-restrictive 26, 221, 278 restrictive 26, 217, 221 resultative construction 238–9, 240 Rhetorical Structure Theory 213 right-detached position 6, 192 status 9–11 structure, relational vs non-relational 3, 4 subjacency 273, (see also relative clauses, constraints on; topicalization, constraints on; questions, WH-questions, constraints on) subject 67, 92, 99, 105, (see also privileged syntactic argument) derived intransitive 97 dummy 130 inverted 74, 104 within NP 26 subordinate clause 184, 186, 214, (see also nexus, subordination; relative clauses) adverbial clause 183, 186 complement 183, 217 fronting 193 switch-reference 106, 183 syntactic inventory 13–16, 24, 130, 171 Systemic Functional Grammar 2 telicity 33–4, 37–8 templates, syntactic 13–16, 24, 130, 171, 223 combining 13–16 selection principles 130, 142, 244, 255, 262 tense 8–11, 186, 201 text 192 that-clause 189, 199 thematic relations 53–61 topic 68, 229 topic chain 103–5, 229, (see also conjunction reduction) topic–comment, see focus, predicate topicalization 273, 285 constraints on 286–8 transitivity M[acrorole]-transitivity 64, 118 S[yntactic]-transitivity 64, 255 Undergoer, see macroroles, semantic

309

310

Subject index

verb phrase (VP) 8, 80, 234 VP ellipsis 231–4, 291 voice 115, 116, 140 antipassive 98, 100, 116, 133, 147, et passim argument modulation 116, 117

passive 8, 60, 100, 115, 117, 119, 132, 140, 159, et passim PSA modulation 116, 164 wa vs ga (Japanese) 73, 83 want constructions, see control construction zero anaphora 173

Exploring the Syntax-Semantics Interface

Exploring the Language of Drama: From Text to Context (Interface)

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

The Machiavelli Interface

The Kruton interface

The Kruton Interface

The Prosody-Morphology Interface

The Machiavelli Interface

The Ni-YSZ interface

The Biology-Chemistry Interface

The immersed interface method

Engineering The User Interface

software interface

software interface

manufacture interface

Exploring the Syntax-Semantics Interface

Exploring the Language of Drama: From Text to Context (Interface)

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

Interface

The Machiavelli Interface

The Kruton interface

The Kruton Interface

The Prosody-Morphology Interface

The Machiavelli Interface

The Ni-YSZ interface

The Biology-Chemistry Interface

The immersed interface method

Engineering The User Interface

software interface

software interface

manufacture interface

Recommend Documents